Testing Modification Sensitivity to Credibility and Ballast

Why Test Sensitivity?

An experience modification translates a risk’s past losses relative to its expected losses into a factor applied to premium. The modification can change when the underlying experience changes, but it can also change when the plan gives that experience more or less weight. Testing sensitivity to credibility and ballast separates these effects: keep actual and expected losses fixed, change one plan input at a time, and compare the resulting modifications.

The comparison is meaningful only when the calculation basis stays consistent. Use the same actual losses, expected losses, exposure period, loss definition, and any plan adjustments in every scenario. The examples below use hypothetical aggregate losses and assume those items have already been put on a consistent basis. No interest or mortality assumptions apply; no claim frequency or severity distribution is assumed.

Definition: Let $A$ be the risk’s actual losses included in the modification, $E$ its expected losses on the same basis, $Z$ a credibility weight between 0 and 1, and $B$ a nonnegative ballast amount. In the credibility form, the modification is $M=1+Z(A/E-1)$, for $E>0$. In the ballast form, it is $M=(A+B)/(E+B)$. These are equivalent when $B=E(1-Z)/Z$ for $Z>0$. A plan’s stated formula governs; do not apply both forms as separate adjustments.

How Credibility and Ballast Relate

In the credibility form, the experience indication $A/E$ is blended with a neutral modification of 1. If $Z=1$, the experience receives full weight and $M=A/E$. If $Z=0$, the experience receives no weight and $M=1$. For $0<Z<1$, the result lies between 1 and $A/E$.

Ballast expresses the same dampening effect in dollars. Adding the same nonnegative amount to actual and expected losses pulls their ratio toward 1. A larger ballast gives the fixed experience less influence. The conversion between the two forms follows by setting the ballast ratio equal to the credibility expression:

\frac{A+B}{E+B} =1+\frac{A-E}{E+B} =1+\frac{E}{E+B}\left(\frac{A}{E}-1\right).

Thus the credibility represented by ballast $B$ is $Z=E/(E+B)$. Solving for $B$ gives $B=E(1-Z)/Z$. Substitution verifies the equivalence: $E+B=E/Z$, so $(A+B)/(E+B)=1+Z(A/E-1)$. This relationship also shows why credibility and ballast are not independent parameters in these equivalent formulas. Choosing a different $Z$ implies a different equivalent $B$, and vice versa.

Important condition: The ratio $A/E$ and the conversion $B=E(1-Z)/Z$ require $E>0$. At $Z=0$, the credibility form gives $M=1$, while its equivalent ballast is the limiting case $B\to\infty$. Follow a plan-specific rule when expected losses are zero; do not use these ratios by dividing by zero.

A Practical Sensitivity Procedure

A sensitivity test is a controlled comparison, not a new estimate of the risk’s losses. Record a baseline modification, change one specified parameter, recalculate, and report the absolute difference. If the plan supplies credibility weights, use the credibility formula. If it supplies ballast amounts, use the ballast formula. When comparing changes stated in different parameterizations, calculate the modification under each stated scenario before judging the size of the change.

Freeze the experience.
Keep $A$, $E$, and the loss and exposure basis fixed across scenarios.

Identify the plan input.
Determine whether the scenario gives $Z$ or $B$, and use the corresponding formula.

Calculate each modification.
Show the baseline and alternative values before rounding, then compute their absolute difference.

Interpret the direction and size.
If $A>E$, more credibility raises $M$, while more ballast moves $M$ downward toward 1. If $A<E$, the directions reverse.

Worked Calculations

Worked Example: Comparing a Credibility Change With a Ballast Change

A hypothetical contractor has actual losses of $A=\$240{,}000$ and expected losses of $E=\$200{,}000$. Compare the modification at credibility $Z=0.25$ with the modification at $Z=0.55$. Separately, compare ballast amounts of $B=\$400{,}000$ and $B=\$200{,}000$. Treat all amounts as fixed aggregate losses on the same basis; no interest, mortality, or loss distribution is assumed.

Method and credibility comparison. First calculate the experience indication: $A/E=240{,}000/200{,}000=1.20$, so $A/E-1=0.20$. At $Z=0.25$, $M=1+0.25(0.20)=1.05$. At $Z=0.55$, $M=1+0.55(0.20)=1.11$. The absolute modification change is $1.11-1.05=0.06$.

Ballast comparison. At $B=\$400{,}000$, $M=(240{,}000+400{,}000)/(200{,}000+400{,}000)=640{,}000/600{,}000=1.0667$. At $B=\$200{,}000$, $M=(240{,}000+200{,}000)/(200{,}000+200{,}000)=440{,}000/400{,}000=1.10$. The absolute change is $1.10-1.0667=0.0333$, rounded to four decimal places.

Conclusion and check. For these specified alternatives, the credibility change produces the larger modification movement: $0.06$ versus approximately $0.0333$. Both results are above 1 because actual losses exceed expected losses. The ballast results move toward 1 as ballast increases, as expected. The two comparisons are separate what-if tests, not independent credibility and ballast adjustments applied together.

Common mistake avoided: Using $A/E=1.20$ as the modification in every scenario ignores the stated credibility or ballast. The correction is to use the parameter specified for each calculation.

Worked Example: Lower-Than-Expected Losses

A hypothetical fleet has $A=\$80{,}000$ and $E=\$100{,}000$. Calculate the modifications at $Z=0.30$ and $Z=0.65$, then compare ballast amounts of $\$300{,}000$ and $\$100{,}000$. Assume fixed aggregate amounts, no interest or mortality, and no assumed probability distribution.

Method and credibility comparison. The experience indication is $80{,}000/100{,}000=0.80$, so its difference from 1 is $-0.20$. At $Z=0.30$, $M=1+0.30(-0.20)=0.94$. At $Z=0.65$, $M=1+0.65(-0.20)=0.87$. The absolute change is $|0.87-0.94|=0.07$.

Ballast comparison. With $B=\$300{,}000$, $M=(80{,}000+300{,}000)/(100{,}000+300{,}000)=380{,}000/400{,}000=0.95$. With $B=\$100{,}000$, $M=(80{,}000+100{,}000)/(100{,}000+100{,}000)=180{,}000/200{,}000=0.90$. The absolute change is $|0.90-0.95|=0.05$.

Conclusion and check. The credibility alternatives change the modification by $0.07$, compared with $0.05$ for the stated ballast alternatives. Since $A<E$, increasing credibility gives the below-1 experience indication more influence and lowers the modification. Increasing ballast moves the result upward toward 1: the $\$300{,}000$ ballast result, $0.95$, is closer to 1 than the $\$100{,}000$ result, $0.90$.

Common mistake avoided: A larger credibility weight does not always increase the modification. It moves $M$ farther from 1 in the direction of $A/E$; here that direction is downward.

Understanding the Size of a Change

The formulas make the direction and size of a one-parameter change explicit. For fixed $A$ and $E>0$, the credibility form can be written as $M=1+Z(A-E)/E$. Therefore, changing credibility from $Z_1$ to $Z_2$ changes the modification by:

M(Z_2)-M(Z_1)=(Z_2-Z_1)\left(\frac{A}{E}-1\right).

The absolute change is the product of the absolute credibility change and the absolute experience deviation from 1. If $A=E$, the experience indication is exactly 1 and changing credibility has no effect. If losses differ substantially from expected, the same change in credibility has a larger effect.

For ballast, write the modification as $M=1+(A-E)/(E+B)$. For two ballast values $B_1$ and $B_2$, direct subtraction gives:

M(B_2)-M(B_1) =(A-E)\left(\frac{1}{E+B_2}-\frac{1}{E+B_1}\right) =\frac{(A-E)(B_1-B_2)}{(E+B_1)(E+B_2)}.

This expression shows that increasing ballast reduces the absolute deviation from 1 when $A\ne E$, provided the denominators are positive. The result does not say that a dollar of ballast has the same impact as a particular change in credibility. Their numerical scales differ, so compare the resulting modification changes for the actual alternatives being tested.

Worked Example: Checking a Ballast Change Against Its Credibility Equivalent

A hypothetical repair business has $A=\$360{,}000$ and $E=\$300{,}000$. Compare credibility $Z=0.50$ with $Z=0.65$. Then compare the ballast equivalent of $Z=0.50$, namely $\$300{,}000$, with an alternative ballast of $\$240{,}000$. Use fixed aggregate values; assume no interest, mortality, or loss distribution.

Credibility calculation. Here $A/E=360{,}000/300{,}000=1.20$, so the deviation from 1 is $0.20$. At $Z=0.50$, $M=1+0.50(0.20)=1.10$. At $Z=0.65$, $M=1+0.65(0.20)=1.13$. The absolute change is $0.03$.

Ballast calculation. For $Z=0.50$, the equivalent ballast is $B=300{,}000(1-0.50)/0.50=\$300{,}000$. Its modification is $(360{,}000+300{,}000)/(300{,}000+300{,}000)=660{,}000/600{,}000=1.10$, checking the equivalence. At $B=\$240{,}000$, $M=600{,}000/540{,}000=1.1111$. The ballast comparison therefore changes the modification by $1.1111-1.10=0.0111$, rounded to four decimal places.

Conclusion and check. The specified credibility change has the larger effect: $0.03$ versus approximately $0.0111$. Also, $B=\$240{,}000$ corresponds to credibility $Z=E/(E+B)=300{,}000/540{,}000=0.5556$, which is higher than $0.50$ and produces a modification between 1.10 and 1.13. The direction and equivalent-weight check are consistent.

Common mistake avoided: Do not add ballast to actual losses but leave expected losses unchanged. Ballast is added to both sides of the ratio; otherwise the calculation no longer represents the stated ballast form.

Common Interpretation Traps

Changing the experience while testing a parameter. If actual losses, expected losses, or the loss basis also change, the difference cannot be attributed solely to credibility or ballast.
Comparing parameter values instead of modification changes. A change of 0.10 in $Z$ cannot be directly compared with a dollar change in $B$. Calculate the modifications first.
Applying both equivalent forms at once. The ballast formula already represents credibility through its relationship to $Z$. Applying a separate credibility adjustment again double-counts the effect unless a plan explicitly specifies a different model.
Assuming all changes move $M$ upward. More credibility moves the modification away from 1 toward $A/E$; the direction depends on whether actual losses are above or below expected.
Rounding too early. Retain intermediate values at full precision, then round the final modification or difference as requested. Small changes can otherwise be obscured.

Key takeaway: For a fixed loss experience, calculate each scenario under its stated formula and compare absolute modification changes. Credibility and ballast are equivalent representations when $B=E(1-Z)/Z$; they should not be treated as two independent adjustments unless the plan defines them that way.

Check Your Understanding

For each question, keep the stated experience fixed and use the indicated formula. Report modification differences to four decimal places when needed.

A hypothetical risk has $A=\$126{,}000$, $E=\$120{,}000$, and $Z=0.35$. Calculate the modification. (Answer: $1.0175$)
A hypothetical account has $A=\$90{,}000$, $E=\$120{,}000$, and $B=\$180{,}000$. Calculate its ballast-form modification. (Answer: $0.9000$)
For fixed $A=\$210{,}000$ and $E=\$175{,}000$, calculate the modifications at $Z=0.20$ and $Z=0.60$, then give the absolute change. (Answer: $1.0400$, $1.1200$; change $0.0800$)
For fixed $A=\$150{,}000$ and $E=\$200{,}000$, compare $B=\$100{,}000$ with $B=\$300{,}000$. Which ballast gives the modification closer to 1, and what is the absolute change? (Answer: modifications $0.8333$ and $0.9000$; $B=\$300{,}000$ is closer to 1; change $0.0667$)
A risk has $A=\$260{,}000$, $E=\$200{,}000$, and baseline credibility $Z=0.40$. Compare its credibility-form modification with the ballast-form modification at $B=\$150{,}000$. Identify the larger of the two modifications and their absolute difference. (Answer: $Z=0.40$ gives $1.1200$; $B=\$150{,}000$ gives $1.1714$; the ballast-form result is larger by $0.0514$)

What You'll Learn