Arithmetic Versus Geometric Trend Averages

Why the Choice of Average Matters

Annual trend rates describe percentage changes, and percentage changes compound. If a hypothetical insurance loss measure increases by 10% in one year and then decreases by 10% the next, the changes do not cancel: the second change applies to the increased amount. Consequently, an average of the annual rates is not automatically the rate that reproduces the total change over several years.

The arithmetic average summarizes the rates themselves. The geometric average is the constant annual rate that gives the same cumulative change as the observed sequence. For a multi-year projection intended to preserve a known cumulative trend, that distinction is essential. The calculations below treat each rate as applying over one full year to the measure at the start of that year.

Definition: Let $r_1,\ldots,r_n$ be annual trend rates for $n$ successive, equal-length annual periods, with each $r_j>-1$. The arithmetic average rate is $\bar r_A=(r_1+\cdots+r_n)/n$. The cumulative trend factor is $F=\prod_{j=1}^n(1+r_j)$. The geometric average rate is the constant annual rate $r_G$ satisfying $(1+r_G)^n=F$; thus $r_G=F^{1/n}-1$.

The condition $r_j>-1$ ensures every annual growth factor $1+r_j$ is positive, so the geometric average is defined. A negative annual trend rate is permitted as long as it is greater than $-100\%$. The number of periods $n$ must count the actual annual intervals represented by the rates.

Why the Geometric Average Reproduces the Endpoint

Suppose the measure at the beginning of the first year is $L_0$. After the first year it is $L_0(1+r_1)$. After the second year, the second year's rate applies to that new amount, giving $L_0(1+r_1)(1+r_2)$. Continuing in this way gives the endpoint after $n$ years:

L_n=L_0\prod_{j=1}^n(1+r_j)=L_0F.

If a single constant annual rate $r_G$ is to give the same endpoint, applying it for $n$ years must produce $L_0(1+r_G)^n=L_0F$. Dividing by $L_0$, when $L_0>0$, and taking the positive $n$th root gives:

1+r_G=F^{1/n} \qquad\Longrightarrow\qquad r_G=\left[\prod_{j=1}^n(1+r_j)\right]^{1/n}-1.

This is a direct consequence of compounding: the constant rate is chosen to match the product of the annual growth factors, not the sum of the annual rates. It reproduces the same total change over the specified period, though it does not claim that each actual year had that constant rate.

Result: For positive annual growth factors, the geometric average of the growth factors is no greater than their arithmetic average. Since the arithmetic average of the factors $1+r_j$ is $1+\bar r_A$, this means $r_G\leq\bar r_A$, with equality when all annual rates are equal. This comparison concerns these two averages of the same annual rates; it does not imply that the arithmetic average is never useful.

Worked Calculations

Worked Example: A Three-Year Projection With Rising Rates

A hypothetical loss measure is $2,000 at the start of a three-year period. Its annual trend rates are 10%, 21%, and 33.1%. Calculate the arithmetic average, the cumulative factor, the geometric average, and the three-year projection using each average as a constant rate. State which projection matches the actual sequence.

Method and rates. There are three annual intervals, so $n=3$. The annual growth factors are $1.10$, $1.21$, and $1.331$.

Arithmetic average. Add the rates and divide by three:

\bar r_A=\frac{0.10+0.21+0.331}{3} =\frac{0.641}{3} =0.2136666667\ldots

Thus the arithmetic average is 21.3667% when expressed to four decimal places as a percentage.

Cumulative factor and geometric average. Multiply the annual growth factors. In this example the factors are $1.1$, $1.1^2$, and $1.1^3$, so their product is $1.1^6=1.771561$. The geometric average is:

F=1.10(1.21)(1.331)=1.771561, \qquad r_G=(1.771561)^{1/3}-1=1.21-1=0.21.

The geometric average is 21%. To check, $1.21^3=1.771561$, exactly the cumulative factor.

Project the measure. Using the actual sequence, the projected amount is $2000(1.771561)=\$3{,}543.122$, or $3,543.12 rounded to the nearest cent. Using the geometric average gives the same result. Using the arithmetic average instead gives:

2000\left(1+\frac{0.641}{3}\right)^3 =2000(1.7877149526\ldots) =\$3{,}575.43.

The arithmetic-average projection is $32.31 higher than the projection under the actual sequence. The geometric average is the appropriate constant rate when the aim is to reproduce the three-year endpoint. The common mistake is to treat the arithmetic mean of the rates as if it preserved compounding.

Worked Example: Mixed Positive and Negative Annual Trends

A hypothetical claim-cost index has annual trend rates of 3%, −2%, and 8% over three successive years. Find the arithmetic average, cumulative factor, and geometric average. If the starting index amount is $5,000, compare the endpoint under the observed rates with a projection that applies the arithmetic average for all three years.

Method and arithmetic average. The number of annual periods is $n=3$. The arithmetic average is:

\bar r_A=\frac{0.03-0.02+0.08}{3} =\frac{0.09}{3}=0.03=3\%.

Cumulative factor. Retain the negative rate as a negative change, so its growth factor is $1-0.02=0.98$:

$$ F=(1.03)(0.98)(1.08) =(1.0094)(1.08) =1.090152. $$

Geometric average. The constant annual rate with this cumulative effect is:

r_G=(1.090152)^{1/3}-1 \approx 0.0291903 =2.9190\%.

As a check, $1.0291903^3$ is approximately $1.090152$; the small difference is from displaying the rate to seven decimal places as a decimal. The actual endpoint from a $5,000 starting amount is $5000(1.090152)=\$5{,}450.76$. The geometric-average projection has the same endpoint before rounding.

Applying the arithmetic average of 3% for each year instead gives:

5000(1.03)^3 =5000(1.092727) =\$5{,}463.635 \approx \$5{,}463.64.

The arithmetic-average projection is $12.88 higher than the endpoint from the observed sequence. The trap is to assume that a negative year can simply be added to positive years and then treated as if all years had the resulting average rate. The correct cumulative calculation multiplies each year's growth factor.

Worked Example: Equal and Opposite Rates Do Not Cancel

A hypothetical repair-cost measure rises by 20% in the first year and falls by 10% in the second year. Its initial value is $10,000. Calculate the arithmetic and geometric averages, the actual two-year endpoint, and the endpoint obtained by applying the arithmetic average in both years.

Arithmetic average. The mean of the two annual rates is:

\bar r_A=\frac{0.20-0.10}{2}=0.05=5\%.

Compounded change and geometric average. The annual growth factors are $1.20$ and $0.90$. Their product is $1.08$, so the measure has increased by 8% overall. The geometric average satisfies:

F=(1.20)(0.90)=1.08, \qquad r_G=\sqrt{1.08}-1 \approx 0.0392305=3.9230\%.

Applying the actual rates to $10,000 gives $10000(1.20)=\$12{,}000$, then $12000(0.90)=\$10{,}800$. This agrees with $10000(1.08)=\$10{,}800$. Applying the geometric average for two years gives the same endpoint because $(1+r_G)^2=1.08$.

Applying the arithmetic average for both years instead gives:

10000(1.05)^2 =10000(1.1025) =\$11{,}025.00.

This is $225.00 above the actual endpoint. A 20% increase followed by a 10% decrease does not return to the starting value, and the arithmetic average does not encode the sequence's cumulative effect. The geometric average does.

Choosing and Reporting the Average

The right calculation depends on the question. If asked for the arithmetic average of annual rates, add the rates and divide by the number of rates. If asked for a constant annual trend that gives the same total change across the period, multiply the growth factors and take the $n$th root. In ratemaking work, these answer different questions and should not be substituted for one another without explanation.

Question	Calculation	What it represents
What is the arithmetic mean of the annual rates?	$\bar r_A=\sum_{j=1}^n r_j/n$	The average of the listed annual rates
What constant annual rate reproduces the total change?	$r_G=[\prod_{j=1}^n(1+r_j)]^{1/n}-1$	The annual rate equivalent to the observed compound change
What is the projected endpoint under the observed rates?	$L_0\prod_{j=1}^n(1+r_j)$	The result of applying the rates successively

These formulas assume the listed rates apply successively to the same measure over equal one-year periods. If the periods are not annual, or are not the same length, first establish the relevant period lengths and an appropriate annualized rate; do not treat unequal intervals as equal annual observations. Also keep the time horizon aligned: an annualized geometric rate derived from $n$ years must be applied for $n$ years to reproduce the cumulative factor.

Avoid premature rounding. Calculate with the full rates and factors available, then round the final reported rate or dollar amount according to the requested precision. Rounding an average before exponentiation can change a multi-year projection, particularly when the starting amount is large or the period is long.

Key takeaway: The arithmetic average describes the average of the annual rates. The geometric average describes the constant annual rate that preserves their cumulative compounded effect. To match a multi-year endpoint, use the product of annual growth factors, not the arithmetic average of the rates.

Check Your Understanding

For each problem, identify which average is requested and account for compounding before rounding.

A hypothetical index has annual rates of 2%, 4.04%, and 6.1208%. Find the arithmetic average, cumulative factor, and geometric average. (Answer: arithmetic average $4.0536\%$; cumulative factor $1.126162419264$; geometric average $4.04\%$.)
A $12,000 hypothetical loss measure is subject to annual trends of 5%, −3%, and 7%. Find its endpoint under the observed sequence and under a constant rate equal to the arithmetic average. (Answer: observed endpoint $=\$13{,}077.54$; arithmetic average $=3\%$; arithmetic-average projection $=\$13{,}112.72$.)
A hypothetical measure increases 25% in one year and decreases 20% in the next. Find the arithmetic and geometric averages, then compare both two-year projections from a starting value of $9,000. (Answer: arithmetic average $=2.5\%$, giving $\$9{,}455.63$; geometric average $=0\%$, giving the actual endpoint $\$9{,}000.00$.)
A hypothetical index has annual trends of 4%, 8.16%, and 12.4864%. Find both average rates and the cumulative factor. If the starting index is 7,500, give the endpoint reproduced by the geometric average. (Answer: arithmetic average $=8.2154667\%$; geometric average $=8.16\%$; cumulative factor $=1.265319018496$; endpoint $=\$9{,}489.89$.)

Question	Calculation	What it represents
What is the arithmetic mean of the annual rates?	\(\bar r_A=\sum_{j=1}^n r_j/n\)	The average of the listed annual rates
What constant annual rate reproduces the total change?	\(r_G=[\prod_{j=1}^n(1+r_j)]^{1/n}-1\)	The annual rate equivalent to the observed compound change
What is the projected endpoint under the observed rates?	\(L_0\prod_{j=1}^n(1+r_j)\)	The result of applying the rates successively

What You'll Learn