Rounding Multiplicative Rating Calculations

Why Rounding Matters in a Multiplicative Rating Plan

A multiplicative indicated rate is found by multiplying a base rate by the relativities for the risk’s characteristics. Even when each factor is rounded by only a small amount, the factor errors combine in the product. The resulting difference may be immaterial in one calculation but important when rates are used repeatedly, compared, or reproduced in a rating system.

The central distinction is between display precision and calculation precision. Displaying a factor as 1.235 does not require replacing its stored value with 1.235 in subsequent calculations. Unless a rating specification requires intermediate rounding, a sound calculation practice is to retain the available precision through the multiplication and round the indicated dollar rate only at the end.

Definition: For a base rate $B$ and $k$ multiplicative relativities $r_1,\ldots,r_k$, the full-precision indicated rate is $R=B\prod_{j=1}^{k}r_j$. If the factors used in the calculation are rounded values $\tilde r_j$, the resulting rate is $\tilde R=B\prod_{j=1}^{k}\tilde r_j$. The rounding difference is $\tilde R-R$; its sign indicates whether the rounded calculation is higher or lower.

Unless stated otherwise, the examples treat all inputs as hypothetical, assume the base rate and factors are given without uncertainty, and use ordinary rounding to the nearest stated decimal place. Dollar amounts are rounded to the nearest cent only when reporting final rates. No interest or mortality assumptions are involved.

Compare the Two Calculation Policies

A factor table may show relativities to a specified number of decimals. That presentation alone does not establish whether the underlying calculation must use those displayed values or higher-precision values. Follow the stated rating procedure. If it does not require intermediate rounding, preserve precision in the calculation and round the final indicated rate.

The following steps make the comparison explicit. The first calculation uses the stated full-precision factors; the second replaces each factor with its rounded value. If the procedure also rounds cumulative products, that is a separate source of rounding and should be identified separately.

Calculate the unrounded product.
Multiply the base rate by all available-precision factors, without rounding intermediate results.

Apply the specified factor rounding.
Round each factor to the required precision, then multiply those rounded factors by the same base rate.

Compare on a common basis.
Subtract the unrounded indicated rate from the rounded-factor rate. Round each final dollar amount to cents for reporting, but do not use those rounded display amounts to calculate the underlying difference.

Worked Example: Rounding Three Relativities

A hypothetical coverage has a base rate of $1,000.00. Its three unrounded relativities are 1.2344, 0.8764, and 1.0004. Compare the indicated rate using the exact stated factors with the rate obtained by rounding every factor to three decimal places. Use the same base rate in both calculations.

Method. Calculate both products independently and round only the final dollar rates to cents. The factors rounded to three decimals are 1.234, 0.876, and 1.000.

Full-precision calculation. First multiply the first two factors:

1.2344 \times 0.8764 = 1.08182816.

Now multiply by the third factor:

1.08182816 \times 1.0004 = 1.082260891264.

Thus the unrounded rate is $1{,}000 \times 1.082260891264 = \$1{,}082.260891264$, which is $1,082.26 to the nearest cent.

Rounded-factor calculation. Multiply the rounded relativities:

1.234 \times 0.876 = 1.080984,\qquad 1.080984 \times 1.000 = 1.080984.

The resulting rate is $1{,}000 \times 1.080984 = \$1{,}080.984$, or $1,080.98 to the nearest cent. Using unrounded dollar amounts, the rounded-factor calculation is lower by $1{,}080.984 - 1{,}082.260891264 = -\$1.276891264$, approximately $1.28 lower.

Check and common mistake. The rounded factors are not all in the same direction relative to their original values, so it would be wrong to assume that rounding must raise or must lower the result. The computed rate is lower here, consistent with the product of the rounded factors being smaller. The comparison is checked by multiplying each set of factors separately by the same $1,000 base.

How Factor Errors Combine

The multiplication itself explains why small factor changes can accumulate. Dividing the rounded rate by the full-precision rate cancels the common base rate, provided $B\ne0$. What remains is a product of factor-by-factor ratios. When the factors are positive, the rate’s relative change is therefore controlled by the combined relative changes in the factors, not by any one factor in isolation.

For example, suppose all factors are rounded upward by the same fraction $e$, so that $\tilde r_j=r_j(1+e)$. The ratio of rounded to exact rates is then $(1+e)^k$, not $1+ke$ exactly. The approximation $1+ke$ can be useful for very small $e$, but the power expression includes the interaction terms and gives the exact relative change under that assumption.

Result: Suppose $B>0$, there are $k$ exact positive factors $r_j$, each $r_j\ge L>h>0$, and each rounded factor satisfies $|\tilde r_j-r_j|\le h$. Then the relative absolute error in the rate is bounded by $\left|\tilde R/R-1\right|\le(1+h/L)^k-1$.

Justification. Because $r_j\ge L$, the absolute rounding condition gives $-h/r_j\le(\tilde r_j-r_j)/r_j\le h/r_j$, and $h/r_j\le h/L$. Hence $1-h/L\le\tilde r_j/r_j\le1+h/L$. The lower endpoint is positive because $L>h$. Since $B>0$ and all exact factors are positive, $R>0$, and

\frac{\tilde R}{R} =\frac{B\prod_{j=1}^{k}\tilde r_j}{B\prod_{j=1}^{k}r_j} =\prod_{j=1}^{k}\frac{\tilde r_j}{r_j}.

Consequently, $(1-h/L)^k\le\tilde R/R\le(1+h/L)^k$. The upward deviation is at most $(1+h/L)^k-1$. The downward deviation is at most $1-(1-h/L)^k$, which is no larger than the upward bound: for $x=h/L$ between zero and one, $(1+x)^k+(1-x)^k\ge2$, by pairing the nonnegative even-power terms in the binomial expansions. This proves the stated absolute-error bound. The condition $B>0$ matters: if $B=0$, both rates are zero and their relative error is undefined.

Worked Example: A Conservative Bound for Upward Rounding

A hypothetical rating calculation has a base rate of $6,075.00 and four exact positive relativities. Each exact factor is at least 0.800, and each is rounded upward by no more than 0.0005. The exact product of the four factors is known to be no greater than 2.000. Give a conservative upper bound on the dollar increase caused by factor rounding, rounded upward to whole cents.

Method. Bound the relative increase in the product of the factors, then apply it to an upper bound for the exact rate. Here $L=0.800$, $h=0.0005$, and $k=4$. Each rounded factor is at most its exact value times $1+h/L$, so the factor product increases by no more than

\left(1+\frac{0.0005}{0.800}\right)^4-1 =(1.000625)^4-1 =0.0025023447267\ldots.

The exact rate is at most $6{,}075 \times 2.000=\$12{,}150.00$. Thus a conservative dollar bound is $12{,}150 \times 0.0025023447267\ldots = \$30.403\ldots$, approximately. Rounded upward to whole cents, the bound is $30.41.

Check and common mistake. The bound uses the largest permitted exact rate, not the base rate by itself. It is also rounded upward to cents because the question asks for a conservative bound; rounding $30.403 down to $30.40 would not guarantee that the reported bound is large enough.

Rounding Cumulative Products Is a Separate Choice

A calculation can round the factors, the cumulative product after each multiplication, or both. These procedures need not give the same answer. Rounding every cumulative product can discard precision more than once, even when the individual factors themselves are not rounded. Therefore a calculation should state exactly what is being rounded and at what stage.

Worked Example: Comparing Factor and Cumulative-Product Rounding

Use the same hypothetical base rate of $1,000.00 and factors 1.2344, 0.8764, and 1.0004. Compare (a) carrying all precision until the final rate, (b) rounding each cumulative relativity to three decimals while retaining the original factors, and (c) rounding both the factors and each cumulative relativity to three decimals. In each rounded procedure, use the rounded cumulative relativity for the next multiplication.

(a) No intermediate rounding. The factor product is 1.082260891264, so the rate is $1,082.260891264, or $1,082.26.

(b) Round cumulative products only. The first cumulative product is $1.2344 \times 0.8764=1.08182816$, which rounds to 1.082. Then $1.082 \times 1.0004=1.0824328$, which rounds to 1.082. The indicated rate is therefore $1{,}000 \times 1.082=$ $1,082.00, $0.26 below the full-precision rate when final dollar amounts are compared.

(c) Round factors and cumulative products. Round the first factor to 1.234. The next factor is retained as 0.8764 for multiplication, so $1.234 \times 0.8764=1.0814776$, which rounds to 1.081. Round the final factor to 1.000; then $1.081 \times 1.000=1.081$. The rate is $1,081.00, $1.26 below the full-precision rate using the displayed final dollar amounts.

Check and common mistake. The three rates differ because the procedures do not use the same intermediate values. In particular, it would be incorrect to describe procedure (b) as “rounding the factors”: its factors remain at full precision, and only the cumulative products are rounded. The multiplication sequence above verifies each result.

Practical Checks and Exam Traps

Rounding rules should be applied consistently, but they should not be invented. A problem may explicitly instruct that factors be rounded before use, or that each cumulative relativity be rounded to a stated precision. Follow that instruction. If it only asks for a final dollar rate, carrying full precision and rounding the final answer is generally the clearest calculation.

Do not round twice without instruction. Rounding factors and then rounding cumulative products introduces two distinct sources of rounding.
Do not add factor errors as if that were exact. Relative errors multiply. For small errors, adding their percentages may approximate the result, but the exact product includes cross-terms.
Keep the base rate consistent. A comparison of rounding policies must use the same base rate in both calculations.
Distinguish a signed difference from its magnitude. State whether rounding raised or lowered the rate, and report the absolute difference if asked for a size.
Do not treat a rounded display as a calculation instruction. A table’s number of displayed decimals does not, on its own, specify how the underlying rate must be computed.

Key takeaway: A multiplicative rating calculation compounds factor values, so rounding choices can affect the final indicated rate. Identify the required rounding stage, retain precision elsewhere, and compare the resulting rates using unrounded calculations before reporting dollars to cents.

Check Your Understanding

For each problem, follow the stated rounding policy and round final dollar amounts to cents unless the question requests a different form.

A hypothetical risk has a base rate of $500.00 and exact factors 1.246 and 0.934. Compute the indicated rate without intermediate rounding. (Answer: $581.88)
A hypothetical equipment policy has a base rate of $2,400.00 and factors 1.0826 and 0.9734. If each factor is rounded to three decimals before multiplication, what rate results? (Answer: $2,529.02)
A calculation’s exact indicated rate is no greater than $1,200. Each of three positive factors is at least 0.950 and is rounded to the nearest 0.001. Using the relative-error bound, give a conservative dollar bound on the absolute change, rounded upward to cents. (Answer: $1.90)
A hypothetical base rate is $800.00, with factors 1.1116 and 0.9044. Round each factor to three decimals, multiply, and round only the final dollar rate. (Answer: $804.20)
Explain why a relative-error formula comparing rounded and exact indicated rates requires a nonzero base rate. (Answer: The relative error divides by the exact rate; if the base rate is zero, both rates are zero, so that ratio is undefined.)

What You'll Learn