9.1 Chain ladder and Bornhuetter-Ferguson methods

Chain Ladder and Bornhuetter-Ferguson Methods

Chain ladder and Bornhuetter-Ferguson are two foundational techniques for estimating ultimate claims in insurance. Actuaries use them to project how much an insurer will eventually pay out on claims that have already occurred, which directly determines how much money needs to be held in reserve.

Both methods build on claims development triangles, but they differ in a critical way. Chain ladder relies exclusively on historical claims data, while Bornhuetter-Ferguson blends historical data with an external estimate of expected losses (an a priori loss ratio). This distinction matters most for recent origin periods where limited data makes pure historical projection unreliable.

Chain Ladder Method

The chain ladder is a deterministic reserving technique that projects ultimate claims by extrapolating observed development patterns forward. Its core logic is straightforward: if claims in a given line of business have historically grown by predictable proportions from one development period to the next, you can use those proportions to fill in the unknown portions of the triangle.

Assumptions of Chain Ladder

• Claims will continue to develop in each period as they have historically.
• The proportional increase in cumulative claims from one development period to the next is the same across all origin periods.
• Claims are fully developed after a finite number of development periods (this number varies by line of business).
• There have been no material changes in claims handling processes, policy benefits, or reinsurance arrangements that would distort historical patterns.

If any of these assumptions break down, chain ladder estimates can become unreliable quickly.

Development Factors in Chain Ladder

Development factors (also called age-to-age factors or link ratios) measure how much cumulative claims grow from one development period to the next. You calculate each factor as:

$f_{j,j+1} = \frac{C_{i,j+1}}{C_{i,j}}$

where $C_{i,j}$ is the cumulative claims for origin period $i$ at development period $j$.

You compute these ratios for every origin period that has data at both development periods $j$ and $j+1$, then select a single factor for each development period transition. Common selection methods include:

• Simple average of the individual factors
• Volume-weighted average (sum of numerators divided by sum of denominators), which gives more weight to larger origin periods
• Excluding outliers or unusual years before averaging

The selected factors form a chain that you multiply together to project from any point in the triangle to ultimate.

Calculating Ultimate Claims with Chain Ladder

1. Organize cumulative claims into a development triangle, with origin periods as rows and development periods as columns.
2. Calculate age-to-age factors for each adjacent pair of development periods.
3. Select a single development factor for each transition (using averaging or judgment).
4. For each origin period, multiply the latest observed cumulative claims by the product of all remaining selected factors:

$\hat{C}_{i,n} = C_{i,k} \times f_{k,k+1} \times f_{k+1,k+2} \times \cdots \times f_{n-1,n}$

where $k = n - i + 1$ is the latest observed development period for origin period $i$, and $n$ is the final development period.

1. Sum the projected ultimate claims across all origin periods to get the total.

Advantages vs Disadvantages of Chain Ladder

• Advantages:
  • Simple, intuitive, and widely used across the industry
  • Requires minimal inputs: just historical cumulative claims data
  • Produces estimates quickly once the triangle is set up
• Disadvantages:
  • Highly sensitive to changes in development patterns over time
  • Does not incorporate exposure or premium information
  • Assumes the future will resemble the past
  • Can produce volatile or unreliable estimates for the most recent origin periods, where very little data has emerged

Bornhuetter-Ferguson Method

The Bornhuetter-Ferguson (BF) method addresses a key weakness of chain ladder: its instability for immature origin periods. BF does this by anchoring the estimate of unreported claims to an a priori expected loss, rather than relying entirely on the small amount of emerged data. As more claims data comes in, the method naturally shifts weight toward actual experience.

Assumptions of Bornhuetter-Ferguson

• The a priori loss ratio is a reasonable estimate of expected claims as a percentage of earned premium.
• Actual emerged claims experience is credible and should influence the ultimate estimate.
• The proportion of ultimate claims that have emerged by a given development period is consistent across origin periods.
• No material changes in business mix, claims handling, or policy benefits have occurred.

A Priori Loss Ratio in Bornhuetter-Ferguson

The a priori loss ratio represents the expected claims as a percentage of earned premium for a given line of business, before looking at the actual emerged claims for that origin period. Sources for setting it include:

• Historical loss ratios from mature origin periods
• Pricing assumptions used when the business was underwritten
• Industry benchmarks or regulatory data

For example, if the a priori loss ratio is 70% and earned premium for an origin period is $1{,}000$, the expected ultimate claims for that period would be $700$.

Expected vs Actual Claims in Bornhuetter-Ferguson

Expected claims for each origin period are calculated as:

$E_i = P_i \times LR$

where $P_i$ is the earned premium for origin period $i$ and $LR$ is the a priori loss ratio.

Actual claims are the cumulative claims that have been reported (or paid, depending on the triangle basis) for each origin period as of the valuation date.

The BF method uses expected claims to estimate the unreported portion and actual claims for the reported portion. This is the key difference from chain ladder, which uses actual claims to drive both portions.

Calculating Ultimate Claims with Bornhuetter-Ferguson

1. Determine the proportion of ultimate claims reported at each development period. You can derive these from the chain ladder cumulative development factors:

$q_{j} = \frac{1}{f_{j,j+1} \times f_{j+1,j+2} \times \cdots \times f_{n-1,n}}$

This gives the reciprocal of the cumulative development factor from period $j$ to ultimate, representing the fraction of ultimate claims that have emerged by period $j$.

1. Calculate the expected unreported claims for each origin period:

$UERC_i = E_i \times (1 - q_{k})$

where $q_{k}$ is the proportion reported for origin period $i$ at its latest observed development period $k$.

1. Estimate ultimate claims as the sum of actual reported claims and expected unreported claims:

$\hat{C}_{i,n} = C_{i,k} + UERC_i$

Notice that the unreported portion depends on the a priori expected claims, not on the actual emerged claims. This is what gives BF its stability for recent origin periods.

Advantages vs Disadvantages of Bornhuetter-Ferguson

• Advantages:
  • More stable estimates for immature origin periods, since the unreported portion is anchored to the a priori expectation
  • Incorporates external information (premium, loss ratios) that chain ladder ignores
  • Responsive to actual experience as it emerges, because reported claims feed directly into the ultimate estimate
• Disadvantages:
  • Results are sensitive to the selected a priori loss ratio; a poor choice propagates through all estimates
  • Assumes the proportion reported is consistent across origin periods
  • May not fully capture shifts in development patterns over time
  • Requires more inputs and judgment than chain ladder

Comparison of Methods

Similarities of Chain Ladder and Bornhuetter-Ferguson

• Both rely on historical claims development patterns to project ultimate claims.
• Both assume that claims will continue to develop in a manner consistent with historical experience.
• Both organize claims data into a development triangle and use development factors (or derived proportions reported).
• Both produce deterministic point estimates of ultimate claims.

Differences Between Chain Ladder and Bornhuetter-Ferguson

Criteria for Selecting Chain Ladder vs Bornhuetter-Ferguson

• Maturity of the business: Chain ladder works well for mature lines with stable development. BF is preferred for newer lines or recent origin periods with sparse data.
• Data stability: If historical claims data is volatile or lacks credibility, BF's external anchor provides more reliable estimates.
• Availability of external information: BF requires a credible a priori loss ratio. If reliable benchmarks or pricing data aren't available, chain ladder may be the more practical choice.
• Desired responsiveness: Chain ladder reacts more quickly to actual claims experience. BF dampens that responsiveness, which is a strength when data is thin but a weakness when genuine shifts in experience need to be captured.

In practice, actuaries often run both methods and compare the results. Large divergences between the two signal that either the development patterns or the a priori assumptions (or both) deserve closer scrutiny.

Reserving Applications

Estimating IBNR with Chain Ladder and Bornhuetter-Ferguson

IBNR (Incurred But Not Reported) represents claims that have occurred but have not yet been reported to the insurer as of the valuation date. It's the primary quantity that reserving methods aim to estimate.

• Chain ladder IBNR: The difference between projected ultimate claims and currently reported claims:

$IBNR_i = \hat{C}_{i,n} - C_{i,k}$

• Bornhuetter-Ferguson IBNR: Equal to the expected unreported claims:

$IBNR_i = E_i \times (1 - q_k)$

Notice that BF's IBNR estimate for a given origin period does not depend on the actual reported claims for that period. This is a direct consequence of how the method is constructed.

Setting Reserves Using Chain Ladder and Bornhuetter-Ferguson

Reserves represent the total amount an insurer must hold to cover future claim payments. For each origin period:

$Reserves_i = IBNR_i + \text{Case Reserves}_i$

where case reserves are the estimated future payments on claims that have already been reported but not yet fully settled. Total reserves are the sum across all origin periods.

Both methods use the same reserve formula; they differ only in how IBNR is calculated.

Monitoring Claims Development

Once reserves are set, ongoing monitoring is essential:

1. Actual vs. expected analysis: Compare actual claims development each period to what the model predicted. Significant deviations should be investigated to determine whether they reflect random variation or a genuine change in the underlying process.
2. Update estimates: As new data emerges, recalculate development factors, reassess the a priori loss ratio, and re-estimate ultimate claims. Quantify the impact of new data on reserve levels.
3. Validate assumptions: Regularly test whether the key assumptions still hold. For chain ladder, check whether age-to-age factors remain stable across origin periods. For BF, assess whether the a priori loss ratio is still appropriate given recent experience.

Limitations and Considerations

Data Requirements

• Sufficient volume and history of claims data to establish credible development patterns
• Consistent claims reporting and settlement practices over time
• Accurate and complete data, including claim counts, amounts, reported/paid dates, and case reserves
• Appropriate segmentation by line of business, coverage type, or other relevant factors to ensure homogeneity within each triangle

Adjustments for Unusual Claims

Large or catastrophic claims can distort development patterns and require special treatment:

• Exclude them from the main triangle and estimate their ultimate cost separately.
• Alternatively, apply a large claims loading to the projected ultimate claims.

Changes in the operating environment (new claims handling procedures, policy benefit changes, reinsurance restructuring) can also invalidate historical patterns. When these occur, you may need to adjust historical data to make it comparable to the current environment, or use judgment to override mechanically calculated factors.

Sensitivity Testing

Because both methods depend on assumptions, sensitivity testing is critical:

• For chain ladder: test alternative age-to-age factor selections and tail factors.
• For BF: test alternative a priori loss ratios and proportion-reported assumptions.
• Identify which assumptions have the largest impact on the estimates and focus your validation efforts there.
• Consider presenting a range of estimates based on reasonable alternative assumptions rather than a single point estimate.

Uncertainty in Estimates

Both methods produce point estimates, but actual outcomes will differ. Techniques for quantifying that uncertainty include:

• Bootstrapping: Resample residuals from the development triangle to generate a distribution of possible ultimate claims outcomes.
• Stochastic modeling: Incorporate probability distributions for development factors or loss ratios to simulate a range of future claims paths (e.g., the Mack model for chain ladder, or stochastic BF variants).

Communicating this uncertainty to stakeholders is part of the actuary's role. Reserves are typically set to include a margin for adverse development, and actual experience should be monitored against estimates on an ongoing basis to trigger adjustments when needed.