What is Chain Ladder Method (CLM) and How It Works

Criptolog » Finance » What is Chain Ladder Method (CLM) and How It Works

The chain ladder method (CLM) is one of the classic actuarial tools used to estimate how much an insurer will ultimately have to pay for claims that have already occurred, including those not fully reported yet. It is especially popular in non-life insurance lines such as property, casualty and health, because it takes historical claims development patterns and uses them to project future development in a simple but powerful way. When people ask “What is the chain ladder method CLM?”, they are usually looking for a clear explanation of how this technique estimates unpaid claims and IBNR (incurred but not reported) reserves.

At its core, CLM works by organizing claims data into development triangles, calculating development factors (also known as age-to-age factors or link ratios), and then projecting each accident year’s losses forward to an ultimate value. The method is widely used because it is relatively straightforward, transparent, and responds quickly to emerging experience; but that same responsiveness can become a drawback in volatile lines or when recent data is distorted. Understanding both how it works and when it can mislead is crucial for actuaries, risk managers and anyone who relies on insurance financial statements.

What is the Chain Ladder Method (CLM)?

The chain ladder method, sometimes called the development method, is an actuarial reserving technique used to estimate ultimate losses and IBNR based purely on historical development patterns observed in claims data. It looks at how claims for past accident periods have grown from one valuation date to the next, assumes those patterns will broadly continue, and applies them to more recent accident years that have not yet fully developed.

In practice, CLM is applied to triangles of either paid losses or incurred losses (paid plus case reserves), organized by origin period and development period. Origin periods are typically accident years, underwriting years, or policy years, while development periods correspond to the age of the claim (for example 12, 24, 36 months from occurrence or from policy inception). Each cell in the triangle represents the cumulative amount of losses for a given origin year evaluated at a specific age.

Insurers are required to hold claim reserves on their balance sheets, and the chain ladder method provides a systematic way to estimate those reserves from past data. The ultimate losses projected by CLM, minus the amounts already reported or paid, represent the unpaid claims liabilities, including IBNR. These figures directly influence reported profit, solvency, and capital requirements.

Because it is simple, data-driven and widely taught, CLM has become one of the most frequently used reserving techniques in actuarial practice. It is a standard benchmark method and is often compared against other approaches such as the Bornhuetter-Ferguson technique and the Expected Loss Ratio method to check consistency and reasonableness of reserve estimates.

Key Concepts Behind the Chain Ladder Method

Before getting into the mechanics, it helps to understand a few key concepts that appear again and again when discussing the chain ladder method. These include run-off or reserve triangles, paid versus incurred losses, loss development factors, cumulative development factors, and tail factors.

Run-off triangles (also called delay triangles or reserve triangles) are two-dimensional matrices that accumulate claim data by origin period and development age. The rows usually represent accident years (1998, 1999, 2000, etc.), and the columns show evaluation lags such as 12, 24, 36, 48 months and so on. Each entry in the triangle shows the cumulative amount of loss for that origin year at that particular age.

For example, in a typical incurred loss triangle, a value like 43,169,009 at the cell for accident year 1998 and 24 months means “cumulative reported losses as of 24 months for claims that occurred in 1998.” As time goes on, more claims are reported and existing claims develop, so the cumulative amounts usually grow across the row until they reach a plateau that approximates ultimate losses.

The chain ladder method can be applied either to paid losses only, or to incurred losses (paid plus case reserves), depending on the insurer’s data quality and reserving philosophy. Paid triangles focus on hard cash flows but may be more delayed, while incurred triangles incorporate case reserves and respond sooner but depend on claims handling practices.

Loss development factors (LDFs), also known as age-to-age factors or link ratios, measure how losses grow from one valuation age to the next. For instance, if cumulative reported losses for a given accident year are 43,169,009 at 24 months and 45,568,919 at 36 months, the age-to-age factor from 24 to 36 months is 45,568,919 / 43,169,009 ≈ 1.056. A factor greater than 1 indicates that losses are still developing.

Cumulative development factors (CDFs) are created by multiplying selected age-to-age factors along the development path to ultimate. Once you have a CDF from a given age to ultimate, you can project an origin year’s ultimate losses simply by multiplying its latest observed cumulative loss by the corresponding CDF.

A tail factor is an extra development factor applied beyond the last observed development column of the triangle. It accounts for any additional development expected after the final evaluation age in the data set. In some portfolios, development after, say, 10 years is negligible and the tail factor can be close to 1.000; in others (e.g., long-tail liability), a material tail must be estimated.

Assumptions Underlying the Chain Ladder Method

The central assumption of the chain ladder method is that the historical pattern of loss development is a reasonable guide to future development. In simple terms, the way claims have grown from one year to the next in the past is assumed to be similar to how they will grow going forward, unless there is clear evidence to the contrary.

For that assumption to hold, the claims data used must be reliable and consistent across time. Significant changes in the insurer’s operations, product mix, coverage terms, legal environment, or claims handling can distort the observed development patterns and break the key premise of CLM. If that happens, the raw chain ladder output may need explicit adjustment.

Examples of changes that can undermine the method include shifts in claims settlement speed, major modifications to case reserving practices, reorganization of the claims department, or regulatory reforms that affect claim frequency or severity. Periods with unusually severe claims, catastrophe events, or one-off shocks can also produce atypical development factors that should not be projected into the future without judgment.

Another practical issue is data volume and volatility. When there are only a few origin years, or when claim sizes are highly erratic, individual age-to-age factors can fluctuate wildly from one year to the next. In such cases, simple averages of the factors may be unstable, and the chain ladder method can produce very noisy estimates unless the actuary smooths or adjusts the data.

Because of these sensitivities, actuaries often supplement company-specific triangles with broader industry data, especially when local experience is thin or distorted. Blending internal and external information can help stabilize development patterns and make the CLM output more credible, though it also introduces model risk if the industry patterns are not fully representative of the company’s business.

Step-by-Step Application of the Chain Ladder Method

Several authors, including Jacqueline Friedland in “Estimating Unpaid Claims Using Basic Techniques,” lay out the chain ladder procedure as a sequence of clear steps. While there are small variations in practice, the standard approach can be summarized in seven main stages that transform raw claim data into ultimate loss estimates.

1. Compile claims data into a development triangle. The first job is to organize the losses into a matrix where rows represent origin periods (for example accident years 1998-2007) and columns represent valuation ages (12, 24, 36, 48, 60 months, and so on). Each entry should be cumulative for that origin year at that age, using either paid or incurred basis consistently.

2. Calculate age-to-age factors (link ratios). For each development interval (12-24, 24-36, 36-48 months, etc.), compute the ratio of the cumulative losses at the later age to those at the earlier age, by accident year. For example, if a year’s reported losses move from 38,954,484 at 12 months to 46,045,718 at 24 months, the 12-24 factor is 46,045,718 / 38,954,484 ≈ 1.182.

3. Compute averages of the age-to-age factors. Once you have a column of factors for each development interval, you usually summarize them by taking simple averages (over the last 3 or 5 years, for example) or volume-weighted averages based on the underlying loss amounts. The idea is to smooth out random noise and obtain representative development factors.

4. Select claim development factors. The actuary then exercises judgment to choose a final factor for each development interval, informed by multiple averaging methods, trends in the factors, and qualitative knowledge of the portfolio. You might, for instance, lean on a 5-year volume-weighted average but adjust slightly if there is evidence of gradual change.

5. Determine the tail factor. If the triangle does not run all the way to ultimate, you need a tail factor from the last observed development age to full maturity. Sometimes this is set to 1.000 when later development is negligible; in other cases, especially for long-tailed lines, the tail factor is estimated from extended data, industry benchmarks, or explicit modeling of late development.

6. Calculate cumulative claim development factors. For each column (each development age), multiply together the selected age-to-age factors from that age through to ultimate, including the tail factor if any. These products form the cumulative development factors, which tell you how much more development is expected from the current age to the final settlement level.

7. Project ultimate claims. Finally, apply the cumulative development factors to the latest observed cumulative losses for each origin period. For example, if the most recent reported loss amount for accident year 2006 is 54,641,339 and the CDF from that age to ultimate is 1.110, the projected ultimate for that year is 54,641,339 × 1.110 ≈ 60,651,886. Do this for each row of the triangle and then sum across rows to obtain total ultimate losses.

Once the ultimate losses are projected, incurred but not reported (IBNR) is obtained by subtracting current reported amounts from those ultimate projections. In a worked example, if total reported losses across all accident years sum to 543,481,587 and total ultimate losses estimated by CLM are 569,172,456, the implied IBNR is 569,172,456 − 543,481,587 = 25,690,869. This amount would appear on the insurer’s balance sheet as part of the outstanding claims liability.

Building and Interpreting the Development Triangle

The starting point of any chain ladder analysis is constructing the data triangle correctly, because errors or inconsistencies here will flow straight through to the reserve estimate. Each claim payment or reported amount must be mapped to two time dimensions: origin and development.

The origin period (row) is usually the accident year, defined by the date on which the insured event occurred. In some contexts, particularly in certain lines of business, the origin period might instead be the underwriting year (policy inception date) or the reporting year (when the claim was first notified). The choice must be applied consistently, as it affects how patterns are interpreted.

The development period (column) captures valuation age, such as 12, 24, 36 months since the origin date. For each claim, the actual calendar date of payment or reporting is converted into the appropriate development age bucket. The triangle entries usually represent cumulative amounts at each age, not incremental changes, because cumulative data tends to be more stable.

Looking at a real data set, you might see a row for accident year 1998 with cumulative reported losses progressing from 37,017,487 at 12 months to 43,169,009 at 24 months, then to 45,568,919 at 36 months and so on. Over time, each additional column fills in for newer accident years, while older years move closer and closer to ultimate, eventually showing almost no further development.

Once the triangle is in place, the next step is to compute the age-to-age factor triangle, which is essentially a triangle of ratios between successive columns. For example, the 12-24, 24-36, 36-48 and later intervals are represented as separate columns of factors for each accident year, except for the latest years which may not yet have enough development to calculate all ratios.

From Age-to-Age Factors to Cumulative Development

After constructing the age-to-age factor triangle, actuaries summarize the factors in each development column using different averaging methods to reduce random variation. Common options include simple averages over the most recent 3 or 5 years, as well as volume-weighted averages where each year’s factor is weighted by the earlier age’s loss amount.

For instance, suppose the last five accident years yield 12-24 month factors around 1.168, and the last three years yield 1.164. Similarly, for the 24-36 interval, averages might come out near 1.058 (5-year) and 1.056 (3-year). These averages are placed into a summary table, often alongside volume-weighted versions, to make it easier to compare and select final assumptions.

The actuary then selects a final set of age-to-age factors, often leaning on the more stable averages but with judgment to reflect any perceived trends or anomalies. For example, the selected series might be: 1.164 from 12-24 months, 1.056 from 24-36, 1.027 from 36-48, 1.012 from 48-60, 1.005 from 60-72, 1.003 from 72-84, 1.002 from 84-96, 1.001 from 96-108, 1.000 from 108-120, plus a tail factor of 1.000 to ultimate.

To get cumulative development factors, each selected age-to-age factor is multiplied by the factors for all subsequent intervals, including the tail factor. This yields a decreasing series of CDFs as you move to older ages, such as 1.292 at 12 months, 1.110 at 24 months, 1.051 at 36 months, 1.023 at 48 months, 1.011 at 60 months, 1.006 at 72 months, 1.003 at 84 months, 1.001 at 96 months and 1.000 thereafter.

Multiplying the latest cumulative losses for each accident year by the corresponding CDF gives you the projected ultimate losses by year. For instance, accident year 2005 might have the latest reported loss of 56,786,410 at a development age where the CDF to ultimate is 1.051, yielding a projected ultimate of about 59,682,517. Repeating this for all years and summing across the triangle provides total ultimate claims for the portfolio.

As a final step, subtracting current reported (or paid) totals from these ultimate projections yields unpaid claims, which can then be broken down by accident year and analyzed further. This breakdown is particularly useful when actuaries need to explain reserve movements, justify assumptions to management, or identify which origin years are driving changes in the total reserve.

Strengths and Limitations of the Chain Ladder Method

One of the biggest strengths of the chain ladder method is its transparency and ease of communication. Stakeholders can see exactly how historical patterns are being used, how the factors are calculated and selected, and how those choices translate into ultimate loss projections and IBNR. This makes CLM a natural reference point when comparing different reserving methods.

Another advantage is that CLM is highly responsive to new information. As fresh data flows into the triangles, development factors can be updated and ultimate estimates will move accordingly. This reactivity is very useful for lines where claims evolve fairly smoothly and where recent experience is a good predictor of the near future.

However, that same responsiveness can become a weakness when experience is extremely volatile, distorted by one-off events, or subject to rapid change in operations. In such cases, the chain ladder method can overreact to unusual years or outliers unless the actuary deliberately excludes or adjusts those factors before averaging.

The method also has no built-in mechanism to incorporate external information about expected loss ratios, pricing assumptions, or exposure measures. It relies entirely on past patterns in the observed data. That means it can be unreliable for new lines of business, new products, or situations where claims development is still immature and historical experience is scarce or unrepresentative.

Actuaries therefore need to apply judgment throughout the process: checking stability of age-to-age factors, assessing operational or environmental changes, and occasionally removing obviously anomalous factors before finalizing selections. For example, if a single accident year shows a wildly high development factor due to a large, late-reported claim, that factor may be excluded from the averaging base to avoid skewing future projections.

Comparison with the Bornhuetter-Ferguson Method

The Bornhuetter-Ferguson (BF) method is often discussed alongside chain ladder because both are widely used for loss reserving but they take quite different approaches to incorporating prior expectations. Understanding the contrast helps clarify when a pure chain ladder approach might not be the best choice.

While the chain ladder method relies solely on historical development patterns, BF combines those patterns with an independent view of the ultimate loss level, often based on an expected loss ratio applied to premium or exposure. Essentially, CLM asks, “If the past repeats itself, what should the ultimate be?”, whereas BF asks, “Given what we expected to happen, and what we have seen so far, how much of the ultimate should we recognize now?”

In practice, CLM works best when the data is rich, stable and credible, particularly for mature portfolios with long, consistent history. BF, on the other hand, shines when early triangle data is thin or unreliable—such as for very recent accident years, new lines of business, or situations with structural breaks—because it tempers the impact of volatile actual data with a more stable a priori expectation.

For example, in a new product launch where only a year or two of claims experience is available, a pure chain ladder might swing wildly as a few claims are reported, whereas a BF approach can anchor the estimate around a pricing-based expected loss ratio. As more credible data accumulates, the BF method gradually gives more weight to actual development and behaves more like a traditional CLM.

In many reserving exercises, actuaries run both methods, comparing pure chain ladder results with BF outputs and using the comparison to challenge assumptions, explain differences, and ultimately pick a best-estimate reserve. Where chain ladder and BF materially diverge, it often signals that further investigation is needed into data quality, operational changes or exposure assumptions.

Comparison with the Expected Loss Ratio Method

The Expected Loss Ratio (ELR) method offers another alternative to chain ladder, especially when claims data is very immature or almost non-existent. Instead of looking at how past claims have developed, ELR starts from a forward-looking view of expected losses per unit of exposure, typically derived from pricing, underwriting judgment or market benchmarks.

Under ELR, the actuary applies an expected loss ratio to earned premium (or a similar exposure measure) to obtain an estimate of ultimate losses, and then subtracts claims already reported or paid to derive reserves. The key point is that the method does not depend on observed development factors and will deliver the same ultimate estimate regardless of how claims have developed so far.

This forward-looking nature makes ELR particularly useful when there is almost no claims triangle to analyze, such as in the very first periods after launching a new product, entering a new territory, or undergoing a major change in policy terms. In these cases, CLM cannot be applied meaningfully because there are no stable patterns to extrapolate.

In contrast, the chain ladder method needs a sufficiently long and reasonably stable history of development to be reliable, and its estimates will move up or down as more claims emerge. That dynamic updating is valuable once the portfolio matures, but it can be misleading at the early stages when a small number of claims could cause big swings.

In practice, reserving teams often use ELR for the very youngest accident years, chain ladder for more mature years where the pattern is credible, and sometimes Bornhuetter-Ferguson as a bridge method in between. The three methods complement each other, and the choice among them depends on data credibility, line of business volatility, and the availability of robust prior expectations.

The chain ladder method remains a cornerstone of actuarial reserving, but it is most powerful when used alongside other techniques and with a clear understanding of its assumptions, strengths and limitations. When historical development patterns are stable and data quality is high, CLM can provide a very informative, responsive view of unpaid claims; when those conditions do not hold, blending it with methods like BF and ELR generally leads to more robust and defensible reserve estimates.