Measuring Rounding and Implementation Effects

Why Rounding Can Change the Realized Rate Change

A selected rate change is often carried into a rating system through factors or premium amounts that must be displayed at limited precision. Even if each individual rounding difference looks small, its effect across a portfolio can be meaningful. The key is to measure the effect against the premiums to which the changes apply, rather than to treat the rounded figures as if they were the original selection.

This tutorial focuses on the deterministic difference between a selected calculation and its implementation. It assumes the same policies, current premiums, and coverage basis are used in both calculations. There is no claim or premium uncertainty in the arithmetic examples: interest and mortality are not applicable, and each calculation is deterministic (equivalently, its distribution is concentrated at the stated value). The methods are aligned with the CAS Exam 5 (Basic Ratemaking and Estimated Claim Liabilities) syllabus.

Definition: Let $P_j$ be current premium for group $j$, $g$ a common selected multiplier, $f_j$ the selected factor for group $j$, and $\tilde f_j$ the factor actually implemented after rounding. The selected aggregate premium is $\sum_j P_j g f_j$, and the implemented aggregate premium is $\sum_j P_j g\tilde f_j$. The implementation effect is implemented premium minus selected premium.

A positive effect means that implementation produces more aggregate premium than the selected calculation; a negative effect means it produces less. The comparison isolates rounding only when all other inputs are held fixed. If the policy population, exposures, coverage, or premium basis also changes, those effects must not be attributed to rounding.

Measuring Factor-Rounding Effects

For each group, the dollar effect is its current premium multiplied by the common multiplier and by the difference between the implemented and selected factors. Summing these group-level effects gives the portfolio effect:

\Delta =\sum_j P_j g(\tilde f_j-f_j) =g\sum_jP_j(\tilde f_j-f_j).

To express the effect as a difference in aggregate rate change, divide by total current premium. The selected and implemented aggregate rate changes are calculated on the same current-premium base:

r_{\text{selected}} =\frac{\sum_jP_jgf_j}{\sum_jP_j}-1, \qquad r_{\text{implemented}} =\frac{\sum_jP_jg\tilde f_j}{\sum_jP_j}-1, \qquad r_{\text{implemented}}-r_{\text{selected}} =\frac{\Delta}{\sum_jP_j}.

The final difference is in rate-change units. Multiplying it by 100 expresses it in percentage points. This is distinct from the percentage difference relative to selected aggregate premium, which is $\Delta/\sum_jP_jgf_j$. State the denominator whenever reporting a percentage effect.

Worked Example: Two Rounded Class Factors

A hypothetical portfolio has current premiums of $90,000 and $210,000. A common multiplier of 1.04 is applied to selected class factors of 1.025 and 0.985. The rating system implements factors rounded to two decimal places: 1.03 and 0.99. Find the dollar effect and the change in the realized aggregate rate change.

Method and assumptions. Compare selected and implemented aggregate premiums on the same $300,000 current-premium base. The calculation is deterministic; there is no interest adjustment or mortality assumption, and no probability distribution is needed.

Selected calculation. For the first class, the full multiplier is $1.04(1.025)=1.066$, so its selected premium is $90{,}000(1.066)=\$95{,}940$. For the second class, the full multiplier is $1.04(0.985)=1.0244$, so its selected premium is $210{,}000(1.0244)=\$215{,}124$. Selected aggregate premium is $\$95{,}940+\$215{,}124=\$311{,}064$, an aggregate rate change of $\$11{,}064/\$300{,}000=3.688\%$.

Implemented calculation. The first implemented full multiplier is $1.04(1.03)=1.0712$, giving $90{,}000(1.0712)=\$96{,}408$. The second is $1.04(0.99)=1.0296$, giving $210{,}000(1.0296)=\$216{,}216$. Implemented aggregate premium is $\$96{,}408+\$216{,}216=\$312{,}624$, an aggregate rate change of $\$12{,}624/\$300{,}000=4.208\%$.

Effect and check. The implementation effect is $\$312{,}624-\$311{,}064=\$1{,}560$, an increase of $1{,}560/300{,}000=0.0052$, or 0.52 percentage points in the aggregate rate change. Directly, both factor differences are $+0.005$, so the effect is $1.04[90{,}000(0.005)+210{,}000(0.005)]=\$1{,}560$. The direct calculation agrees with the aggregate-premium subtraction.

Common mistake avoided. Reporting $312 would understate the effect: it incorrectly uses a factor difference of 0.001 rather than 1.04 times the actual 0.005 difference, which is 0.0052 for each dollar of current premium.

Why the Weighted Calculation Is Necessary

The aggregate formula follows by subtracting like from like. For group $j$, selected premium is $P_jgf_j$ and implemented premium is $P_jg\tilde f_j$. Their difference is $P_jg(\tilde f_j-f_j)$. Adding those differences across groups gives the total effect. Dividing by total current premium gives the difference between the two aggregate rate changes because the common denominator and the minus-one terms cancel:

\begin{aligned} r_{\text{implemented}}-r_{\text{selected}} &=\left(\frac{\sum_jP_jg\tilde f_j}{\sum_jP_j}-1\right) -\left(\frac{\sum_jP_jgf_j}{\sum_jP_j}-1\right)\\ &=\frac{g\sum_jP_j(\tilde f_j-f_j)}{\sum_jP_j} =\frac{\Delta}{\sum_jP_j}. \end{aligned}

This also explains why an unweighted average of factor changes is generally not the portfolio answer. A factor error applied to a large current-premium group contributes more dollars than the same error applied to a small group. If changes are of opposite signs, one group’s effect can offset another’s; the premium-weighted sum captures that offset.

Worked Example: Rounding Effects That Partly Offset

A hypothetical portfolio has current premiums of $120,000, $80,000, and $50,000. The common multiplier is 1.06. Selected factors are 0.973, 1.046, and 1.012; each is rounded to the nearest hundredth, producing implemented factors of 0.97, 1.05, and 1.01. Calculate the dollar effect and the difference in aggregate rate change.

Method and assumptions. Use the current-premium-weighted factor differences. The data are deterministic, with no interest or mortality adjustment and no stochastic distribution.

The factor differences $\tilde f_j-f_j$ are $-0.003$, $+0.004$, and $-0.002$. Their premium-weighted sum is $120{,}000(-0.003)+80{,}000(0.004)+50{,}000(-0.002)=-360+320-100=-\$140$. Applying the common multiplier gives $\Delta=1.06(-\$140)=-\$148.40$.

Check using aggregate premiums. Before applying 1.06, selected weighted premium is $120{,}000(0.973)+80{,}000(1.046)+50{,}000(1.012)=116{,}760+83{,}680+50{,}600=\$251{,}040$. Selected aggregate premium is $1.06(\$251{,}040)=\$266{,}102.40$. Implemented weighted premium is $120{,}000(0.97)+80{,}000(1.05)+50{,}000(1.01)=116{,}400+84{,}000+50{,}500=\$250{,}900$. Implemented aggregate premium is $1.06(\$250{,}900)=\$265{,}954.00$. Subtraction gives $-\$148.40$, as before.

Total current premium is $250,000, so the difference in aggregate rate change is $-148.40/250{,}000=-0.0005936$, or $-0.05936$ percentage points. The selected aggregate rate change is $266{,}102.40/250{,}000-1=6.44096\%$; the implemented change is $265{,}954/250{,}000-1=6.3816\%$. Their difference checks.

Common mistake avoided. Averaging the three factor differences without regard to premium would not measure portfolio impact. The calculation weights each difference by the current premium exposed to it.

Rounding Policy Premiums Is a Separate Operation

A rating system may first apply factors and then round each policy’s premium to a permitted currency unit. This creates a separate effect from rounding the factors themselves. If selected policy premiums are $q_k$, and the implemented premium is $R(q_k)$ under a specified rounding rule $R$, the premium-rounding effect is $\sum_k R(q_k)-\sum_kq_k$. Calculate factor rounding and premium rounding separately if both occur; do not count the same implementation step twice.

Worked Example: Rounding Each Policy Premium to Whole Dollars

A hypothetical insurer has 80 identical policies, each with selected annual premium $316.40. Premiums are rounded to the nearest whole dollar using ordinary half-up rounding. Find the total implementation effect and its percentage of selected aggregate premium.

Method and assumptions. Apply the stated rounding rule to each policy, then compare aggregate premiums. The calculation is deterministic; interest and mortality are not applicable, and no probability distribution is assumed.

For each policy, $316.40 is below the $316.50 half-dollar threshold, so it rounds to $316, not $317. Selected aggregate premium is $80(316.40)=\$25{,}312$. Implemented aggregate premium is $80(316)=\$25{,}280$. Therefore the effect is $\$25{,}280-\$25{,}312=-\$32$.

As a percentage of selected aggregate premium, the effect is $(-32/25{,}312)100=-0.126422\ldots\%$, or $-0.1264\%$ to four decimal places. The per-policy decrease is $0.40, and $80(-\$0.40)=-\$32$, which checks the aggregate calculation.

Common mistake avoided. Do not round $316.40 upward to $317: ordinary half-up rounding moves to the next dollar only at $316.50 or above.

Assessing Whether the Effect Is Material

A rounding effect is not material merely because it is nonzero. Assess its size against a stated benchmark, such as the total current premium, selected aggregate premium, a filing or implementation tolerance, or the dollar scale of the rate change. Report both the dollar amount and a clearly defined relative measure when the effect could affect a decision.

A useful bound is available when factors are rounded to the nearest increment $h$. Under ordinary nearest rounding, each factor’s absolute rounding error is at most $h/2$. If $P_j$ are nonnegative and $g$ is nonnegative, then:

|\Delta| =\left|g\sum_jP_j(\tilde f_j-f_j)\right| \le g\sum_jP_j|\tilde f_j-f_j| \le \frac{gh}{2}\sum_jP_j.

This is a worst-case bound, not a prediction that errors will cancel or attain their maximum. With many policies, positive and negative rounding differences may offset, but the direction and size depend on the actual premium weights and rounding pattern. For policy premiums rounded to the nearest dollar, each policy’s absolute rounding difference is at most $0.50; thus for $n$ policies the absolute total effect cannot exceed $0.50n$ dollars. This bound also does not imply that the realized effect is zero on average for a particular portfolio.

Key takeaway: Compare selected and implemented aggregate premiums on the same premium base. Weight factor errors by the premiums they affect, calculate policy-level premium rounding as a distinct step, and judge materiality using a stated denominator or tolerance.

Common Implementation Traps

Using an unweighted factor average. Aggregate dollars depend on both the factor difference and the current premium in each group.
Confusing percentage with percentage points. A $1,000 effect on $200,000 of current premium changes the aggregate rate change by 0.5 percentage points; it is not automatically a 0.5% relative change in selected premium.
Rounding at the wrong stage. Rounding factors and rounding final policy premiums are different operations and can have different effects.
Assuming rounded values always increase premium. A factor or premium can round upward or downward; the aggregate sign must be calculated.
Ignoring the stated convention. Nearest rounding, truncation, and other system rules can produce different implemented premiums, particularly at ties.

Check Your Understanding

For each question, use the stated premium basis and rounding rule; report the requested implementation effect.

A hypothetical portfolio has current premiums of $40,000 and $60,000, a common multiplier of 1.03, and selected factors 1.024 and 0.986. The factors are implemented as 1.02 and 0.99. What is the dollar effect of factor rounding? (Answer: +$82.40.)
There are 125 hypothetical policies, each with selected premium $482.67, rounded to the nearest whole dollar using ordinary half-up rounding. Find the dollar effect and its percentage of selected aggregate premium, to five decimal places. (Answer: +$41.25; approximately +0.06837%.)
Current premiums are $150,000 and $50,000, with a common multiplier of 1.02 and selected factors 1.037 and 0.949. The implemented factors are 1.04 and 0.95. Find the dollar effect and the change in aggregate rate change in percentage points. (Answer: +$510; +0.255 percentage points.)
Three selected policy premiums are $728.25, $614.50, and $905.10. Each is rounded to the nearest whole dollar using ordinary half-up rounding. Find the total effect. (Answer: +$0.15.)
A portfolio has $500,000 of current premium and a common multiplier of 1.05. Factors are rounded to the nearest hundredth. What is the maximum possible absolute dollar effect under the nearest-rounding bound, and the corresponding maximum effect in percentage points of current premium? (Answer: $2,625; 0.525 percentage points.)

What You'll Learn