Deriving the Marginal Effect of an Added Loss

Why Measure the Effect of One More Dollar?

An experience modification responds to the difference between actual and expected losses, but not every dollar of actual loss has the same effect in every account. Under a credibility-weighted formula, the change depends on both the credibility assigned to experience and the expected-loss denominator. A dollar has a larger effect when credibility is higher or expected losses are lower, provided the plan inputs remain fixed.

This tutorial considers an uncapped modification: no minimum or maximum factor is applied. It focuses on a change to actual losses that are included in the rating calculation. If a plan uses primary losses, for example, a dollar of gross claim payment is not necessarily a dollar of rated actual loss.

Definition: Let $A$ be rated actual losses, $E>0$ be expected losses, and $Z$, with $0\leq Z\leq 1$, be the credibility weight applied to actual experience. In the no-ballast form used here, the uncapped modification is $$ M=1+Z\frac{A-E}{E}. $$ The marginal effect of rated actual loss is the change in $M$ per additional dollar of $A$, with $E$ and $Z$ held fixed.

The formula can also be written as $M=1-Z+ZA/E$. This makes the role of credibility visible: $Z$ weights the actual-to-expected loss ratio, while $1-Z$ supplies the complement. When $A=E$, the modification is 1. When $Z=0$, actual losses have no effect on the modification. When $Z=1$, the modification is the actual-to-expected ratio $A/E$.

Deriving the Marginal Effect

Suppose a new amount $\Delta A$ is added to rated actual losses. Assume the expected losses, credibility, rating period, and all other plan inputs remain unchanged. The new actual losses are $A+\Delta A$, so the new modification is obtained by substituting this amount into the same formula.

\begin{aligned} M_{\text{new}}-M_{\text{old}} &=\left(1+Z\frac{A+\Delta A-E}{E}\right) -\left(1+Z\frac{A-E}{E}\right)\\ &=Z\frac{(A+\Delta A-E)-(A-E)}{E}\\ &=Z\frac{\Delta A}{E}. \end{aligned}

The existing losses cancel because the comparison changes only $A$. Dividing the result by $\Delta A$, for $\Delta A>0$, gives the exact change per dollar:

\frac{M_{\text{new}}-M_{\text{old}}}{\Delta A}=\frac{Z}{E}, \qquad \frac{dM}{dA}=\frac{Z}{E}.

This is an exact finite-change result, not an approximation: the modification is linear in $A$ when $E$ and $Z$ are fixed. The derivative notation expresses the same constant slope. Its units are modification points per dollar of rated actual loss. It is not a percentage increase in the modification; the percentage change would depend on the modification before the added loss.

Key result: With a fixed expected-loss denominator and fixed credibility, adding $\Delta A$ dollars of rated actual loss changes the uncapped modification by $Z\Delta A/E$. Each additional rated dollar changes it by $Z/E$. The change is positive for an added loss when $Z>0$.

Worked Applications

Worked Example: Calculate the Change for a Loss Increase

A hypothetical account has expected losses of $240,000, rated actual losses of $276,000, and a credibility weight of 0.35. Assume the uncapped, no-ballast formula applies and all inputs other than actual losses are fixed. Calculate the current modification, then calculate the change if rated actual losses increase by $12,000. Also find the marginal effect of one additional rated dollar.

Method. Use $M=1+Z(A-E)/E$, then apply the finite-change result $ \Delta M=Z\Delta A/E$. This method avoids recalculating the whole modification without identifying which inputs changed.

Current modification. The loss deviation is $A-E=\$276{,}000-\$240{,}000=\$36{,}000$. Its ratio to expected losses is $\$36{,}000/\$240{,}000=0.15$. Applying credibility gives $0.35(0.15)=0.0525$, so $M=1+0.0525=1.0525$.

Change for $12,000. The modification change is $0.35(\$12{,}000/\$240{,}000)=0.35(0.05)=0.0175$. The new modification is $1.0525+0.0175=1.0700$.

Effect per dollar. The marginal effect is $0.35/\$240{,}000=0.000001458333\ldots$ modification points per dollar. Multiplying by $12,000 gives $0.000001458333\ldots(\$12{,}000)=0.0175$, which checks the finite-change calculation. Reported to 4 significant figures, the marginal effect is 0.000001458 modification points per dollar.

Common mistake avoided. Dividing the increase by expected losses but omitting credibility would give $12{,}000/240{,}000=0.05$, overstating the change. The actual-loss ratio change must be multiplied by $Z=0.35$.

Worked Example: Apply the Effect to Rated Primary Loss

A hypothetical plan uses the first $25,000 of each claim as rated primary loss. Expected losses are $80,000, the credibility weight is 0.60, and current aggregate rated primary losses are $66,000. One claim's gross payment rises from $24,999 to $25,000. Calculate the change in the uncapped modification and the modification before and after the increase. Assume no other inputs change.

Method. First determine the change in rated loss, then apply $Z\Delta A/E$. This avoids treating gross claim dollars above the split point as if they all entered the rated loss amount.

The claim's rated primary loss increases from $24,999 to $25,000, so $\Delta A=\$1$. The gross payment also increases by $1, but here that dollar is still below the $25,000 split point and is fully included.

Modification before the increase. The deviation is $\$66{,}000-\$80{,}000=-\$14{,}000$. The ratio is $-\$14{,}000/\$80{,}000=-0.175$. After credibility weighting, $0.60(-0.175)=-0.105$, so the modification is $1-0.105=0.8950$.

Change and new modification. The change is $0.60(\$1/\$80{,}000)=0.0000075$. Therefore, the new modification is $0.8950+0.0000075=0.8950075$. Substitution provides a check: the new deviation is $\$66{,}001-\$80{,}000=-\$13{,}999$, and $1+0.60(-\$13{,}999/\$80{,}000)=0.8950075$.

Common mistake avoided. If the gross claim had already reached the split point, an additional dollar above it would add no rated primary loss. In that case, $\Delta A=0$, so the modification change would also be zero. Always identify the change in the loss measure used by the formula before applying the marginal effect.

Worked Example: Build the Expected-Loss Denominator First

A hypothetical account has two experience components with expected losses of $45,000 and $75,000. Its rated actual losses are $132,000 and its credibility weight is 0.50. Calculate the uncapped modification, the effect of an additional $2,500 of rated actual losses, and the effect per dollar. Treat expected losses and credibility as fixed.

Method. Add the expected-loss components to obtain the denominator, then use that same denominator for both the current modification and the change. This avoids using actual losses as the denominator.

Expected losses and current modification. Total expected losses are $E=\$45{,}000+\$75{,}000=\$120{,}000$. The deviation is $A-E=\$132{,}000-\$120{,}000=\$12{,}000$, and its ratio to expected losses is $\$12{,}000/\$120{,}000=0.10$. The credibility-weighted deviation is $0.50(0.10)=0.05$, so $M=1.05$.

Change for $2,500. The change is $0.50(\$2{,}500/\$120{,}000)=0.50(0.020833333\ldots)=0.010416666\ldots$. The new modification is $1.05+0.010416666\ldots=1.060416666\ldots$, or 1.060 to 3 decimal places.

Effect per dollar. The marginal effect is $0.50/\$120{,}000=0.000004166666\ldots$ modification points per dollar. Multiplying by $2,500 gives $0.010416666\ldots$, matching the change above. The check confirms both the denominator and the credibility factor were applied consistently.

Common mistake avoided. Replacing the expected-loss denominator with the current actual losses would produce $0.50/\$132{,}000$, which is not the slope of the specified modification formula.

Assumptions, Interpretation, and Limits

The result $Z/E$ applies when the added amount changes rated actual losses alone. In particular, expected losses and credibility must be held fixed. If a problem changes exposure, revises expected losses, or recalculates credibility at the same time as actual losses, the change cannot be found from $Z\Delta A/E$ alone: substitute all revised inputs into the plan formula and compare the two modifications.

Likewise, the dollar in the formula must be a dollar of the plan's rated actual loss measure. If actual losses are adjusted by a per-claim split point, an exclusion, or another stated loss basis, determine how that adjustment changes $A$. The gross payment may rise by one dollar while rated $A$ rises by less than one dollar—or not at all.

If a specified plan includes a stabilizing ballast in its denominator, use the plan's full denominator rather than silently treating it as $E$. For a form $M=1+Z(A-E)/(E+B)$, with fixed ballast $B$, the same subtraction gives a change of $Z\Delta A/(E+B)$. The denominator is determined by the stated formula; the no-ballast result $Z/E$ is not interchangeable with it.

A useful interpretation follows directly: doubling credibility doubles the marginal effect, all else equal, while doubling expected losses halves it. This describes sensitivity of the uncapped modification, not the eventual premium impact. Translating a modification change into a premium change is a separate calculation.

Exam trap: Do not confuse the sign of the loss deviation with the direction of the marginal effect. An account may have $A<E$, so its current modification is below 1, but adding rated actual loss still increases the modification by $Z\Delta A/E$ when $Z>0$. Also, do not use a capped or floored result in this calculation; this result concerns the uncapped formula.

Check Your Understanding

For each problem, use the stated uncapped formula and hold other inputs fixed unless the problem says otherwise.

A hypothetical account has $E=\$50{,}000$, $Z=0.30$, and rated actual losses of $54,000. What is its current modification, and how much does it change if rated actual losses increase by $1,000? (Answer: Current modification $=1.0240$; change $=0.0060$.)
A hypothetical account has expected losses of $90,000 and credibility 0.45. What is the modification change for an increase of $2,400 in rated actual losses, and what is the change per rated dollar? (Answer: Change $=0.0120$; marginal effect $=0.000005000$ modification points per dollar.)
A hypothetical account has $E=\$160{,}000$, $A=\$176{,}000$, and $Z=0.25$. Calculate the current modification, the change from an additional $8,000 of rated losses, and the new modification. (Answer: Current modification $=1.0250$; change $=0.0125$; new modification $=1.0375$.)
A hypothetical plan includes the first $20,000 of each claim as rated primary loss. A claim's gross payment rises from $19,998 to $20,000. The account has expected losses of $72,000 and credibility 0.60. What is the modification change? (Answer: Rated loss increases by $2, so the modification change is $0.60(\$2/\$72{,}000)=0.00001667$, to 4 significant figures.)
A hypothetical account combines two experience components with expected losses of $40,000 and $110,000. Its credibility is 0.32. Calculate the modification change when rated actual losses increase by $7,500, using the combined expected losses as the denominator. (Answer: Combined expected losses $=\$150{,}000$; modification change $=0.32(\$7{,}500/\$150{,}000)=0.0160$.)

What You'll Learn