Adjusting Frequency Trends For Portfolio Mix

Why Portfolio Mix Matters

An aggregate claim frequency can change even when no segment’s own frequency changes. The reason is that the portfolio may contain different proportions of lower- and higher-frequency risks in different periods. A comparison of aggregate frequencies therefore combines two effects: changes in frequency within segments and changes in the exposure weights assigned to those segments.

To measure frequency trend apart from mix change, calculate each period’s segment frequencies and combine them using the same reference weights. This is called standardizing to a common mix. The resulting comparison answers a specific question: how would the periods’ frequencies compare if both periods had the chosen reference distribution of risks?

Definition: For segment \(j\) in period \(t\), let \(N_{jt}\) be the claim count and \(E_{jt}\) the earned exposure on a consistent basis. The segment frequency is \(f_{jt}=N_{jt}/E_{jt}\). The segment’s share of total exposure is \(w_{jt}=E_{jt}/\sum_k E_{kt}\). The aggregate frequency is \(F_t=\sum_j w_{jt}f_{jt}\). Given fixed reference weights \(q_j\), where \(q_j\ge 0\) and \(\sum_j q_j=1\), the standardized frequency for period \(t\) is \(F_t^{(q)}=\sum_j q_j f_{jt}\).

A segment could be a territory, vehicle class, coverage type, or another defined risk group. The exposure basis must be consistent across segments and periods; for example, do not combine vehicle-years in one period with policy-years in another unless the exposure definitions are reconciled. The weights are shares of exposure, not shares of claims.

Why the Aggregate Is an Exposure-Weighted Average

The aggregate frequency is total claims divided by total exposure. Since \(N_{jt}=E_{jt}f_{jt}\), summing segment claims and dividing by total exposure gives:

F_t=\frac{\sum_j N_{jt}}{\sum_j E_{jt}} =\frac{\sum_j E_{jt}f_{jt}}{\sum_j E_{jt}} =\sum_j\left(\frac{E_{jt}}{\sum_k E_{kt}}\right)f_{jt} =\sum_j w_{jt}f_{jt}.

This identity explains the mix effect: even if all the segment frequencies \(f_{jt}\) stay fixed, the aggregate can move when the weights \(w_{jt}\) move. Standardization replaces the period-specific weights with the same \(q_j\) in both periods. The difference \(F_1^{(q)}-F_0^{(q)}\) then reflects differences in segment frequencies under that reference mix, rather than a change in the weights.

A standardized trend is not automatically the trend for every segment or for every possible portfolio. It is a comparison conditional on the selected segments, exposure definitions, and reference weights. Where segment trends differ, a different reference mix can produce a different standardized percentage trend.

Worked Example: A Rising Segment Frequency Hidden by Mix Shift

A hypothetical insurer groups a small commercial property portfolio into two occupancy segments. In period 0, the segment frequencies are 0.020 and 0.100 claims per exposure unit, with equal exposure in each segment. In period 1, both segment frequencies are 5% higher, but the period 1 exposure weights are 80% and 20%, respectively. Calculate both observed aggregate frequencies and the period 1 frequency standardized to the period 0 mix. These are empirical counts and exposures; no interest, mortality, or parametric claim-count distribution is assumed.

Step 1: Calculate the period 0 aggregate. The period 0 weights are \(w_{10}=0.50\) and \(w_{20}=0.50\), so:

$$ F_0=0.50(0.020)+0.50(0.100)=0.010+0.050=0.0600. $$

Step 2: Calculate the period 1 segment frequencies. A 5% increase means multiplying each period 0 frequency by \(1.05\):

f_{11}=0.020(1.05)=0.021,\qquad f_{21}=0.100(1.05)=0.105.

Step 3: Calculate the observed period 1 aggregate. Apply the period 1 weights, not the period 0 weights:

$$ F_1=0.80(0.021)+0.20(0.105) =0.0168+0.0210 =0.0378. $$

Step 4: Standardize period 1 to the period 0 mix. Use the equal period 0 weights with the period 1 segment frequencies:

F_1^{(q)}=0.50(0.021)+0.50(0.105) =0.0105+0.0525 =0.0630.

The standardized change is \(0.0630/0.0600-1=0.05\), or a 5% increase. The observed aggregate instead falls from 0.0600 to 0.0378 because the period 1 portfolio has shifted heavily toward the lower-frequency segment. The method avoids mistaking that mix shift for a decline in the segment frequencies.

Standardized Frequency Trends in Practice

To compare two periods, use the same reference weights for both. A common choice is the base-period exposure mix, but another relevant, credible mix may be selected. State the choice explicitly: “standardized to the period 0 mix” is more informative than reporting a standardized trend without naming its reference.

For two periods, the standardized percentage trend is:

T^{(q)}=\frac{F_1^{(q)}}{F_0^{(q)}}-1, \qquad F_t^{(q)}=\sum_j q_j f_{jt}.

This ratio compares two rates constructed with identical weights. It is not the difference between the observed aggregate trend and a separate estimate of mix change; it is a direct comparison on a fixed composition. If a segment has zero reference weight, its frequency does not affect this standardized measure. If a segment has positive reference weight but no usable frequency estimate in a period, the measure cannot be calculated without addressing the missing or inadequate data.

Worked Example: Counts, Exposure Shares, and a Fixed Reference Mix

A hypothetical insurer records the following claim counts and earned exposures for two equipment-protection segments. Segment A has 300 claims from 10,000 exposure units in period 0 and 420 from 12,000 in period 1. Segment B has 120 claims from 2,000 exposure units in period 0 and 80 from 1,000 in period 1. Standardize both periods to weights of 80% for A and 20% for B, then compare that trend with the observed aggregate trend. Counts are treated as empirical; no probability distribution or interest assumption is needed.

Step 1: Calculate the four segment frequencies.

f_{A0}=\frac{300}{10{,}000}=0.030,\qquad f_{B0}=\frac{120}{2{,}000}=0.060,

f_{A1}=\frac{420}{12{,}000}=0.035,\qquad f_{B1}=\frac{80}{1{,}000}=0.080.

Step 2: Calculate each observed aggregate directly from total claims and exposure. Period 0 has \(300+120=420\) claims and \(10{,}000+2{,}000=12{,}000\) exposure units. Period 1 has \(420+80=500\) claims and \(12{,}000+1{,}000=13{,}000\) exposure units:

F_0=\frac{420}{12{,}000}=0.035,\qquad F_1=\frac{500}{13{,}000}=0.0384615\ldots.

The observed percentage change is \(0.0384615\ldots/0.035-1=0.098901\ldots\), or 9.89% to two decimal places.

Step 3: Apply the fixed 80%/20% reference weights.

F_0^{(q)}=0.80(0.030)+0.20(0.060)=0.024+0.012=0.036,

F_1^{(q)}=0.80(0.035)+0.20(0.080)=0.028+0.016=0.044.

The standardized percentage change is \(0.044/0.036-1=0.2222\ldots\), or 22.22% to two decimal places. The standardized result is larger than the observed 9.89% increase because the actual period 1 portfolio has a lower share of segment B, whose frequency is higher in both periods. The method avoids treating the changing exposure distribution as though it were a change in the underlying segment frequencies.

Separating Frequency Change From Mix Change

An exact decomposition makes the two effects visible. For two periods, add and subtract the intermediate quantity \(\sum_j w_{j0}f_{j1}\):

F_1-F_0 =\sum_j w_{j0}(f_{j1}-f_{j0}) +\sum_j(w_{j1}-w_{j0})f_{j1}.

The first term measures frequency change using period 0 weights. The second measures the effect of changing weights, evaluated at period 1 segment frequencies. This decomposition is exact, but its allocation depends on the chosen order of comparison; an alternative decomposition can evaluate the mix effect at period 0 frequencies instead. Standardization is often easier to interpret when the question is specifically how period frequencies compare under one fixed mix.

Worked Example: Quantifying the Mix Contribution

A hypothetical two-segment portfolio has period 0 weights \(w_{10}=0.70\), \(w_{20}=0.30\) and frequencies \(f_{10}=0.020\), \(f_{20}=0.080\). In period 1, the weights are \(w_{11}=0.50\), \(w_{21}=0.50\), and frequencies are \(f_{11}=0.024\), \(f_{21}=0.088\). Decompose the observed aggregate change into the period-0-weighted frequency effect and the period-1-frequency-weighted mix effect. These are empirical rates; no interest, mortality, or count-distribution assumption is used.

Step 1: Find the two observed aggregates.

$$ F_0=0.70(0.020)+0.30(0.080)=0.014+0.024=0.038, $$

$$ F_1=0.50(0.024)+0.50(0.088)=0.012+0.044=0.056. $$

Thus, the total change is \(F_1-F_0=0.056-0.038=0.018\) claims per exposure unit.

Step 2: Calculate the frequency effect at period 0 weights.

$$ 0.70(0.024-0.020)+0.30(0.088-0.080) =0.70(0.004)+0.30(0.008) =0.0028+0.0024 =0.0052. $$

Step 3: Calculate the mix effect at period 1 frequencies.

$$ (0.50-0.70)(0.024)+(0.50-0.30)(0.088) =(-0.20)(0.024)+(0.20)(0.088) =-0.0048+0.0176 =0.0128. $$

Step 4: Check the decomposition. The two contributions sum to \(0.0052+0.0128=0.0180\), exactly the observed aggregate change. Here, the mix shift toward segment 2 contributes more to the aggregate increase than the segment frequency changes do. The decomposition avoids attributing the entire 0.018 change to frequency trend.

Choosing Weights and Avoiding Misinterpretation

The reference mix should answer the business question. Base-period weights are convenient when the goal is to express later experience on the earlier portfolio composition. A current-period mix may be useful when the question concerns the current portfolio. A fixed, representative mix can support comparisons across several periods. Whichever basis is used, report the weights or describe their source.

If segment frequency trends differ, the selected weights matter. For example, a segment that is growing quickly will have more influence when given a larger reference weight. Therefore, a standardized trend is not a universal portfolio constant. It is an exposure-weighted summary for the stated reference mix.

Before interpreting a result, verify that the periods use comparable claim definitions, exposure measures, segment boundaries, and observation maturity. Changes in claim reporting, coverage, data capture, or classification can make nominally identical segments incomparable. Standardizing weights does not correct such changes, nor does it remove within-segment changes in risk composition.

Common exam trap: Do not use period 1 weights to calculate the standardized period 1 frequency when the specified reference is the period 0 mix. In the first example, \(0.80(0.021)+0.20(0.105)=0.0378\) is the observed period 1 aggregate; \(0.50(0.021)+0.50(0.105)=0.0630\) is the period 1 frequency standardized to the period 0 mix. Confusing these quantities changes the answer and the interpretation.

Key takeaway: Calculate segment frequencies first, then apply the same stated exposure weights to every period being compared. The observed aggregate describes the actual mix; the standardized comparison describes frequency under a common mix.

Check Your Understanding

For each problem, use exposure weights rather than claim shares when aggregating segment frequencies.

Two segments have period 0 frequencies 0.030 and 0.090, with weights 0.60 and 0.40. In period 1, the frequencies are 0.033 and 0.099, but the weights are 0.80 and 0.20. Find both observed aggregates and the period 1 frequency standardized to the period 0 mix. (Answer: observed aggregates 0.0540 and 0.0462; standardized period 1 frequency 0.0594, a 10% increase from period 0.)
Segment A records 90 claims from 3,000 exposures in period 0 and 154 claims from 4,400 exposures in period 1. Segment B records 40 from 500 in period 0 and 18 from 200 in period 1. Using fixed reference weights of 75% for A and 25% for B, calculate both standardized frequencies and their percentage change. (Answer: standardized frequencies 0.0425 and 0.04875; increase 14.71%.)
Two segment frequencies remain 0.040 and 0.120 in both periods. The weights change from 75%/25% to 90%/10%. Find the observed aggregate in each period and the standardized period 1 frequency using period 0 weights. What does the comparison show? (Answer: observed aggregates 0.060 and 0.048; standardized period 1 frequency 0.060, so the observed decline is entirely due to mix change.)
Period 0 has weights 0.70/0.30 and frequencies 0.020/0.080. Period 1 has weights 0.50/0.50 and frequencies 0.024/0.088. Using the exact decomposition in the tutorial, calculate the frequency and mix contributions to the aggregate change. (Answer: frequency contribution 0.0052; mix contribution 0.0128; total aggregate change 0.0180.)
Segment frequencies change from 0.020/0.100 to 0.030/0.110. Calculate the standardized percentage trend first using reference weights 0.80/0.20, then using 0.40/0.60. (Answer: 27.78% using 0.80/0.20 weights; 14.71% using 0.40/0.60 weights, illustrating that the standardized trend depends on the chosen reference mix.)

What You'll Learn