Using the wrong loss elimination ratio formula

The loss elimination ratio is E[X ^ d] / E[X], not E[(X-d)+] / E[X]. It represents the proportion of losses eliminated by the deductible, not the proportion retained. The complement gives the expected payment ratio.

Mixing up cumulative and incremental triangles

Age-to-age development factors apply to cumulative triangles. If the data is given as incremental (paid or incurred in each period), you must convert to cumulative before computing development factors.

Forgetting to adjust for censoring and truncation in MLE

When fitting distributions to insurance data, some observations are censored (limited by policy limits) or truncated (missing below the deductible). The likelihood function must account for these data features.

Errors in Buhlmann-Straub exposure weighting

Buhlmann-Straub weights the credibility by exposure (premium, claim count, etc.). Using equal weights when exposures differ produces the basic Buhlmann estimate, which is suboptimal when exposure data is available.

Confusing indicated rate change with indicated rate level

The indicated rate change is the percentage adjustment needed. The indicated rate level is the absolute rate. Computing the change as (indicated - current) / current requires using consistent premium bases.

Not applying trend factors correctly

Loss trend and premium trend may differ. Losses must be trended from the average date of loss to the future effective date. Premiums must be on-leveled to current rate level. Applying both trends to the same quantity is incorrect.

Misinterpreting GLM deviance residuals

Deviance residuals are not the same as raw residuals. They are derived from the contribution of each observation to the model deviance. Using raw residuals in a GLM context ignores the distributional assumptions.

Incorrect VaR and TVaR calculations

VaR at level alpha is the alpha-th quantile of the loss distribution. TVaR (or CTE) is the expected loss given that the loss exceeds VaR. Candidates sometimes compute TVaR as the average of all losses, not just those exceeding VaR.

Confusing accident year and calendar year development

Accident year development tracks claims from the same accident period. Calendar year data aggregates all payments made in a calendar period regardless of accident date. Using the wrong basis misaligns development patterns.

Over-relying on chain-ladder for immature years

Chain-ladder projections for the most recent accident year rely heavily on the selected development factors, especially the tail. For immature years, Bornhuetter-Ferguson or Cape Cod methods are more stable because they anchor to an a priori loss ratio.

Common MAS-II Mistakes