EconStatsCraft › Chain ladder reserving

Chain ladder reserving

Takes a run-off triangle of claims (origin period, development period, amount paid), computes the volume-weighted development factors, projects each origin year to ultimate, and reports the reserve (IBNR) with Mack standard errors and a coefficient of variation.

What it needs

Origin period (accident / underwriting year)
Development period
Claims in that cell

Frequently asked questions

What is the chain-ladder method?

The chain-ladder method projects insurance claims that have occurred but are not yet fully paid. It takes a run-off triangle of cumulative claims by origin year and development year, estimates how much claims grow from one development year to the next (the development factors), and applies those factors to each origin year to reach its ultimate claims.

How do you calculate chain ladder development factors?

For each development year, divide the sum of cumulative claims at the next development year by the sum at the current one, using only origin years observed at both. These volume-weighted factors are then multiplied together from each origin year's latest diagonal to project it to ultimate.

How is the IBNR reserve calculated with chain ladder?

IBNR (incurred but not reported, here the outstanding reserve) is the projected ultimate claims minus the claims already paid on the latest diagonal, origin year by origin year. The total reserve is the sum across origin years.

What are the assumptions of the chain ladder method?

That every origin year develops in the same proportions as past years, so the development pattern is stable. Changes in claims handling, inflation, mix of business or a large one-off claim break that assumption, and the projections for the most recent years, which rest on the longest chain of factors, are the least reliable.

What is the Mack standard error in chain ladder?

Mack's model gives a standard error for the chain-ladder reserve without assuming a distribution for claims. It combines parameter error in the development factors with process variance, so you can quote the reserve with a margin, not just a point estimate.

The intuition

Claims are reported and paid over years, so a recent accident year looks cheap purely because it is young. The chain ladder assumes every year develops in the same PROPORTIONS: if past years grew 1.59 times from development 0 to 1, this year will too. Multiply the latest diagonal out to ultimate with those ratios, subtract what has been paid, and the remainder is the reserve.

When to use it

Setting the outstanding-claims reserve for a general (non-life) insurer: motor, liability, property. It is the workhorse reserving method and the benchmark every other method is compared against.

Where the estimator comes from

Let C_{ij} be cumulative claims for origin year i at development j. The volume-weighted development factor is \hat{f}_j=\frac{\sum_i C_{i,j+1}}{\sum_i C_{ij}}, summed over the origin years observed at BOTH j and j+1. Projected ultimate for a row last seen at development k is \hat{C}_{i,n}=C_{ik}\prod_{j=k}^{n-1}\hat{f}_j, and the reserve is \sum_i(\hat{C}_{i,n}-C_{ik}). Mack (1993) showed these factors are the weighted least squares estimates under the assumptions E[C_{i,j+1}\mid C_{ij}]=f_jC_{ij} and \mathrm{Var}(C_{i,j+1}\mid C_{ij})=\sigma_j^2C_{ij}, which is what makes the volume weighting the right choice and delivers a standard error for the reserve.

The model

f_j = \frac{\sum_i C_{i,j+1}}{\sum_i C_{i,j}},\qquad \hat{C}_{i,J} = C_{i,\,\text{latest}}\prod_{j} f_j,\qquad R_i = \hat{C}_{i,J} - C_{i,\,\text{latest}}

What it assumes

Every origin year follows the same development pattern - no change in claims handling, no shift in case-mix, no inflation shock partway through.
Development factors depend only on the cumulative claims so far, not on the origin year or the calendar year.
The most recent origin years, which rely on the longest projection, are the least certain - and dominate the total reserve.
The tail beyond the triangle is negligible or has been added deliberately. A ten-by-ten triangle assumes development stops at ten periods, and for long-tail business - liability, asbestos - that assumption alone can understate the reserve by more than the triangle contains.

Statistical properties

The method is distribution-free. Mack derived a mean squared error from the two moment assumptions alone, without assuming a Poisson or lognormal model.
The last development factor rests on a single pair of observations, so the tail is the least reliable part of the estimate and often the largest. A fitted tail factor is standard practice.
The method assumes no calendar-year effects. A change in claims handling speed, or inflation, hits every origin year on the same diagonal and breaks the proportionality assumption in a way the residual plot reveals.
It cannot handle a change in mix or in case-reserving philosophy: both look like a change in development pattern and are extrapolated forward as though permanent.
A row with zero or negative incremental development produces a factor below one, which is legitimate for salvage and subrogation but usually signals a data error.
The Bornhuetter-Ferguson method blends the chain ladder ultimate with an a priori loss ratio, and is preferred for immature years where the chain ladder multiplies a small and noisy diagonal by a large factor.

How to read the output

An age-to-age factor of 1.25 means claims grew 25% from one development period to the next; the factors should decline toward 1 as the year matures.
The reserve (IBNR) is the projected ultimate minus what has been paid. The total is the number that goes on the balance sheet.
The Mack standard error is estimation uncertainty, not the full range of outcomes. A coefficient of variation above 20-30% for the total says the triangle is thin or volatile.
The total SE is larger than the square root of the sum of the origin-year variances because the shared development factors correlate the years.

A worked reading

A ten-by-ten paid triangle. The age-to-age factors decay as they should - 1.6021 from development 0 to 1, then 1.2166, 1.1155, 1.0558, 1.0311, 1.0127 - so most of the development happens in the first two years and the tail is thin. Total IBNR comes to 18,338. Look at where it sits rather than at the total: the 2016 origin year needs only 59 more, while 2022 needs 2,609, because the older years are nearly run off and the newest is being multiplied by almost the whole chain of factors. The Mack standard errors tell the other half of the story. In absolute terms uncertainty rises with the reserve (15 for 2016, 332 for 2022), but as a coefficient of variation it FALLS, from 25.9% to 12.7% - the old years have tiny reserves, so a small absolute error is a large relative one. Neither number alone is the risk: a 12.7% CoV on 2,609 is worth more capital than a 25.9% CoV on 59.

Symptom → diagnosis → what to do

If you see this	It means	Do this
Reserve dominated by the newest origin year	Correct: it has the furthest to develop	But check the factor chain applied to it
A development factor below 1	Salvage, subrogation, or a data error	Investigate before accepting it
Residuals trending along the diagonal	A calendar-year effect: inflation or a process change	The core assumption is violated; adjust or use a model that allows it
Tail factor doing most of the work	Too few development periods observed	Fit a curve to the factors rather than stopping
One origin year distorting a factor	A large individual claim	Consider excluding it and reserving for it separately
Very immature latest year	The chain ladder is unstable here	Use Bornhuetter-Ferguson for that row

Common mistakes

Applying it to a triangle with a visible calendar-year effect (an inflation spike, a law change) - the constant-pattern assumption fails and the projection is biased.
Trusting the youngest origin year's ultimate: one or two data points multiplied by a long chain of factors.
Using paid and incurred triangles interchangeably; they develop differently and imply different reserves.
Reading the Mack standard error as the whole uncertainty. It measures the error in the chain-ladder estimate given the method; it says nothing about the risk that the method itself is the wrong one for this book.

Where to read this properly

Mack (1993), "Distribution-Free Calculation of the Standard Error of Chain Ladder Reserve Estimates", ASTIN Bulletin 23(2).
Bornhuetter and Ferguson (1972), "The Actuary and IBNR", PCAS 59.
England and Verrall (2002), "Stochastic Claims Reserving in General Insurance", British Actuarial Journal 8(3).
Wuthrich and Merz, Stochastic Claims Reserving Methods in Insurance (2008).

Related Actuarial

Life table · Life annuity · Life insurance pricing · Buhlmann credibility · Survival analysis · Mortality graduation · Aggregate loss model · Ruin probability · Multi-state Markov model · Loss severity fitting

All analyses

Data & variables

Each example is simulated from a known model, so you can check whether the estimator recovers the truth - and nothing here is anyone's private data.

Filter variables

Drag a variable into a slot on the right - or just click it.

Choose an example dataset above to begin.

Choose a transformation - it opens in the data editor, with a live preview.

Apply to

Multiply by

Columns ctrl-click for several

Group by ctrl-click for several

Within groups

Statistic

Bands

How many

Cut points

Keep rows where

Value

and

Fill with

Value

Also mark which rows were filled

Rules

Base period = 100

Which part

Percent at each tail

Threshold

Survey weight optional

Direction

Names column

Values column

Largest first

Delete them, not just count them

Read this column as

Tidy the text

New name optional

EconStatsCraft rigorous, plain-spoken econometrics

Econometrics that explains itself.

Regression, panel data, time series and the whole battery of tests. Every result is followed by a verdict you can defend.

Load a dataset to see it here.

Chain ladder reserving

What it needs

Frequently asked questions

What is the chain-ladder method?

How do you calculate chain ladder development factors?

How is the IBNR reserve calculated with chain ladder?

What are the assumptions of the chain ladder method?

What is the Mack standard error in chain ladder?

The intuition

When to use it

Where the estimator comes from

The model

What it assumes

Statistical properties

How to read the output

A worked reading

Symptom → diagnosis → what to do

Common mistakes

Where to read this properly

See also

Related Actuarial

All analyses

Charts

Maps

Explore

Regression

Time series - one series

Time series - several series

Panel data

Multivariate statistics

Marketing research

Causal

Finance & risk

Actuarial

Tests & breaks

Machine learning

Econometrics that explains itself.