Why Exposure Adjustment Comes First
A change in claim counts from one experience period to the next does not, by itself, show that claim frequency changed. If the second period has more insured exposure, it may naturally have more claims even when the underlying claim rate is unchanged. Frequency trends therefore compare claims per unit of exposure, not raw claim counts.
This tutorial uses experience periods of one year unless stated otherwise. The claim definition and the basis for counting claims must be consistent from period to period; exposure should cover the same population and time basis as the counts. When those conditions hold, an observed frequency change can be summarized as a trend. It is not automatically evidence that the underlying risk changed: shifts in portfolio mix, claim handling, or data quality can also affect observed frequencies.
The exposure unit must be stated and consistent. For example, if exposure is measured in earned vehicle-years, a frequency of 0.008 means 0.008 claims per earned vehicle-year, or 8 claims per 1,000 earned vehicle-years. A frequency is not the same as the probability of at least one claim: a policy can have more than one claim, so the count-per-exposure measure need not be a probability.
From Successive Frequencies to an Annual Trend
For adjacent annual periods, the observed annual trend is the percentage change in frequency from one year to the next. If frequencies rise from 0.010 to 0.011, the factor is 1.10 and the observed trend is 10%. If the periods are separated by more than one year, the endpoint change must be annualized to express it as an equivalent constant annual rate.
The formula follows by compounding the same annual factor \(1+g\) for each of the \(k\) years. Starting at \(f_0\), after one year the frequency is \(f_0(1+g)\); after two years it is \(f_0(1+g)^2\); continuing in this way gives \(f_0(1+g)^k\). Setting this equal to the observed endpoint \(f_k\), dividing by \(f_0\), and taking the positive \(k\)-th root gives \(1+g=(f_k/f_0)^{1/k}\), hence the stated formula. The frequencies must be positive for this ratio-based calculation.
For several successive periods, one can calculate each adjacent factor \(f_t/f_{t-1}\). Multiplying those factors gives the total endpoint factor because the intermediate frequencies cancel:
Thus the equivalent constant annual trend over the full interval is determined by the endpoints, even when the individual annual changes differ. An arithmetic average of the annual percentage changes generally does not produce that same compounded result.
Worked Example: Comparing Periods With Unequal Exposure
A hypothetical insurer records 84 claims on 12,000 earned vehicle-years in one year and 99 claims on 13,200 earned vehicle-years in the following year. Calculate both annual frequencies and the observed annual frequency trend. Also show why using the count growth alone gives the wrong answer.
Method and assumptions. Use frequency \(f=N/E\), with the same claim definition and earned vehicle-year basis in both periods. The periods are adjacent one-year intervals, so their frequency ratio is the one-year trend factor.
First period frequency. $$ f_1=\frac{84}{12{,}000}=0.007 $$ This is 7 claims per 1,000 earned vehicle-years.
Second period frequency. $$ f_2=\frac{99}{13{,}200}=0.0075 $$ This is 7.5 claims per 1,000 earned vehicle-years.
Frequency trend. The frequency factor is $$ \frac{f_2}{f_1}=\frac{0.0075}{0.007}=1.0714286, $$ so the observed trend is \(1.0714286-1=0.0714286\), or 7.143% rounded to three decimal places.
Common mistake avoided. The raw count rose by \(99/84-1=17.857\%\). That calculation ignores the exposure increase from 12,000 to 13,200, which is 10%. It therefore overstates the frequency change. The final check is that \(0.007\times1.0714286\approx0.0075\), matching the second-period frequency.
Annualizing a Multi-Year Endpoint Change
When the comparison spans multiple years, the endpoint frequency ratio is a cumulative change factor, not an annual trend. Taking the appropriate root converts it to a constant annual factor. This is useful for describing the overall change on a per-year basis, but it does not reveal the path taken between the two endpoints.
Worked Example: Annualizing a Three-Year Frequency Change
A hypothetical specialty insurer observes 900 claims on 100,000 earned policy-years at the beginning of a three-year comparison interval. At the end, it observes 1,215 claims on 120,000 earned policy-years. Find the endpoint frequency factor and the equivalent constant annual trend. Assume consistent claim and exposure definitions throughout.
Method. Calculate each endpoint frequency, take their ratio, and annualize that ratio over the three years.
Beginning frequency. $$ f_0=\frac{900}{100{,}000}=0.009. $$ Ending frequency. $$ f_3=\frac{1{,}215}{120{,}000}=0.010125. $$
Cumulative frequency factor. $$ \frac{f_3}{f_0}=\frac{0.010125}{0.009}=1.125. $$ The ending frequency is 12.5% higher than the beginning frequency over the full three-year interval.
Equivalent annual trend. $$ g=(1.125)^{1/3}-1 \approx 1.0400419-1 =0.0400419. $$ The equivalent constant annual trend is approximately 4.004%, rounded to three decimal places.
Check and common mistake avoided. Compounding the unrounded annual factor gives \(1.0400419^3\approx1.125\), which recovers the endpoint factor. Reporting 12.5% as the annual trend would confuse the three-year cumulative change with a one-year change. The annualized rate is smaller because it compounds over three years.
Reading a Sequence of Annual Changes
A sequence of frequencies can show that the year-to-year experience is uneven. Adjacent changes describe what happened in each interval; the endpoint annualized trend summarizes the full span as if a single constant annual rate had applied. Both are useful, but they answer different questions.
Worked Example: Chain-Linking Successive Periods
A hypothetical coverage program reports the following claim counts and earned exposures in four successive years: Year 1 has 500 claims and 50,000 exposure units; Year 2 has 660 claims and 55,000 exposure units; Year 3 has 780 claims and 60,000 exposure units; Year 4 has 1,331 claims and 100,000 exposure units. Calculate each frequency, the adjacent annual changes, and the equivalent constant annual trend from Year 1 to Year 4. Assume the claim count basis and exposure definition are unchanged.
Step 1: Calculate frequencies. $$ f_1=\frac{500}{50{,}000}=0.010,\quad f_2=\frac{660}{55{,}000}=0.012,\quad f_3=\frac{780}{60{,}000}=0.013,\quad f_4=\frac{1{,}331}{100{,}000}=0.01331. $$
Step 2: Calculate the adjacent changes. The Year 1 to Year 2 factor is \(0.012/0.010=1.20\), a 20% increase. The Year 2 to Year 3 factor is \(0.013/0.012=1.0833333\), an 8.333% increase. The Year 3 to Year 4 factor is \(0.01331/0.013=1.0238462\), a 2.385% increase. These percentages are rounded to three decimal places.
Step 3: Check the chained factor. $$ 1.20\times1.0833333\times1.0238462\approx1.331 \quad\text{and}\quad \frac{f_4}{f_1}=\frac{0.01331}{0.010}=1.331. $$ The adjacent factors therefore agree with the endpoint ratio, subject to rounding.
Step 4: Annualize the three-year endpoint change. $$ g=(1.331)^{1/3}-1=1.10-1=0.10. $$ The equivalent constant annual trend is 10.000%. As a final check, \(1.10^3=1.331\).
Common mistake avoided. The three adjacent percentage changes are not all 10%; the observed changes vary. Their arithmetic mean is approximately \((20+8.333+2.385)/3=10.239\%\), which is not the compounded equivalent annual trend. The endpoint method gives 10% because it solves \(1.331=(1+g)^3\).
Interpretation and Common Traps
A frequency trend is a relative change in claims per exposure unit. Before interpreting it, verify that the numerator and denominator refer to the same periods, that the exposure units are consistent, and that claims are counted on a comparable basis. If the portfolio composition or the way claims are recorded changes, a raw period-to-period frequency comparison may combine several effects. Adjusting for those effects requires additional analysis; the simple calculations here do not do so.
When aggregating counts and exposure across multiple groups or periods, the overall frequency is total claims divided by total exposure, \(\sum N_t/\sum E_t\). It is not generally the unweighted average of the individual frequencies. This matters because periods or groups can have very different exposure volumes.
The most frequent exam trap is to calculate \(N_2/N_1-1\) and call it the frequency trend. That is the change in claim counts. The correct comparison is \((N_2/E_2)/(N_1/E_1)-1\). A second trap is to report a multi-year endpoint percentage change as an annual rate without annualizing it. For a \(k\)-year interval, use the \(k\)-th root of the endpoint factor, not division of the cumulative percentage by \(k\), except as a possible approximation for small changes.
Check Your Understanding
For each problem, use consistent claim counts and exposure units as stated. Report percentage trends to three decimal places where needed.
- A hypothetical insurer records 72 claims on 8,000 earned units in one year and 90 claims on 10,000 earned units the next year. Calculate each frequency and the annual frequency trend. (Answer: 0.009 and 0.009 claims per exposure unit; 0.000% trend.)
- A program has 420 claims on 40,000 exposure units in the first year and 520 claims on 50,000 exposure units in the next year. Calculate the frequency trend. (Answer: approximately \(-0.952\%\).)
- Frequency is 0.008 at the beginning of a two-year interval and 0.00968 at the end. Find the cumulative factor and equivalent constant annual trend. (Answer: factor 1.21; 10.000% per year.)
- Three successive annual frequencies are 0.0200, 0.0190, and 0.02205. Find both adjacent percentage changes and the equivalent annual trend over the full two-year interval. (Answer: \(-5.000\%\), \(16.053\%\), and 5.000% annualized.)
- Claim counts rise from 300 to 360 while exposure rises from 20,000 to 27,000 units. Calculate the raw count growth and the exposure-adjusted frequency trend; explain which one is the trend in frequency. (Answer: count growth 20.000%; frequency falls from 0.015 to 0.013333 claims per unit, a frequency trend of \(-11.111\%\).)