The frequency–severity matrix plots how often a kind of loss happens against how bad one event is. Each of its four boxes points to a first move: retain small rare losses, retain and prevent frequent small ones, transfer and reduce rare large ones, and avoid or restructure frequent large ones. It is built from the loss runs, cause by cause.
What is the frequency–severity matrix?
The frequency–severity matrix (also called the risk management matrix, and related to the Prouty approach in ARM study material, which uses four frequency and three severity levels) is a simple grid that turns a loss history into a first recommendation. Put frequency, how often a kind of loss happens, on one axis. Put severity, how bad one event is, on the other. Split each into low and high, and every exposure lands in one of four boxes.
It belongs to step 2 of the risk management process, analysis, and it hands step 3 a short list of techniques to examine.
What does each box tell you to do?
- Low frequency, low severity: retain and watch. Losses are rare and small. Absorb them in the budget and keep an eye on the trend.
- High frequency, low severity: retain and prevent. Losses are common but small. Insurance is an expensive way to pay for them, so budget for them inside a deductible or retention, and use loss prevention to cut the count.
- Low frequency, high severity: transfer and reduce. Losses are rare but large. This is what insurance is for. Loss reduction, such as fast reporting and return to work, keeps the next one from growing.
- High frequency, high severity: avoid or restructure. Losses are common and large. Change or stop the activity, and expect what remains to need both prevention and insurance.
The two off-diagonal boxes are where the matrix earns its keep. Expected loss is frequency times severity, so a frequent small loss and a rare large loss can carry the same expected cost and still need opposite treatment. The techniques themselves are explained in risk control techniques.
How are frequency and severity measured?
Frequency is counted from the loss runs: claims on a line or cause, divided by the years observed, gives claims a year. To compare a 40-person contractor with a 400-person one, convert it to a rate per 100 full-time workers using the federal convention the Bureau of Labor Statistics publishes: claims × 200,000 ÷ hours worked. The 200,000 is the hours 100 employees work in a year at 40 hours a week for 50 weeks. It is the same base used for TRIR and DART rates.
Severity is the average claim and, more importantly, the largest claim at the latest valuation.
Then each axis needs a threshold, the line between low and high. These are judgment calls and should be written next to the grid:
- A common frequency threshold is one claim a year: if it happens every year, it is frequent.
- A sensible severity threshold is the point where a claim stops being the contractor's money and becomes the carrier's, such as the per-claim deductible. For workers' comp, another natural line is the split point, the claim size above which NCCI stops counting a claim at full weight in the experience mod.
Change a threshold and boxes move. The claims do not.
A worked example (illustrative numbers)
A contractor works 120,000 hours a year. Over two policy years its workers' comp loss runs show 12 strain claims at $5,000 each and 2 falls at $60,000 each. Thresholds: frequency 1 a year, severity $25,000. The figures are illustrative.
Strains
- Claims a year: 12 ÷ 2 = 6
- Rate per 100 workers: 6 × 200,000 ÷ 120,000 = 10.0
- Average and largest: $5,000
- High frequency (6 is at least 1), low severity ($5,000 is under $25,000): retain and prevent
Falls
- Claims a year: 2 ÷ 2 = 1
- Rate per 100 workers: 1 × 200,000 ÷ 120,000 = 1.67
- Average and largest: $60,000
- High frequency at a threshold of 1 a year, high severity: avoid or restructure
Move the frequency threshold to 2 a year and the falls read as low frequency, high severity: transfer and reduce. Nothing about the falls changed. Either way the practical answer is the same pair: insurance carries the $60,000 claim, return to work and fast reporting keep the next one smaller, and prevention effort goes first to the strains, which are 12 of the 14 claims.
Why should you split the matrix by cause?
Look at the whole workers' comp line instead: 14 claims, 7 a year, $180,000 incurred, an average of 180,000 ÷ 14 = $12,857. That average describes no actual claim. It blends twelve $5,000 strains with two $60,000 falls. A deductible chosen on the $12,857 average would be wrong in both directions.
Draw the matrix per line (workers' comp, general liability, auto) and per cause within each line (strains, falls, rear-end collisions). A line that reads "avoid" as a whole is almost always a frequent small cause sitting on top of a rare large one.
What can the matrix not do?
It does not price anything. The box tells you which family of technique to look at; loss forecasting estimates what next year's losses will cost, and cost-benefit analysis decides whether a given control pays. It also depends on clean, developed loss data: open claims that are under-reserved will sit in the wrong box until they mature.
Used well, the matrix keeps a contractor from treating a frequency problem with insurance or a severity problem with a training video.
Quick answers
What is a frequency–severity matrix in risk management?
It is a two-by-two grid that plots how often a kind of loss happens against how large one loss is. Each box suggests a first treatment: retain, retain and prevent, transfer and reduce, or avoid.
Should high-frequency, low-severity losses be insured?
Usually they are better retained inside a deductible and attacked with loss prevention, because insurance on predictable small losses costs the losses plus the carrier's expenses. The right deductible still depends on cash flow and the rest of the program.
How do you calculate claim frequency per 100 workers?
Multiply the number of claims by 200,000 and divide by total hours worked. The 200,000 represents 100 full-time employees working 40 hours a week for 50 weeks, the base BLS and OSHA use.
- Expected loss
- Frequency times severity: how many losses a year multiplied by the average cost of one. Six $5,000 claims a year is $30,000 of expected loss.
- Split point
- The dollar line that divides each claim into primary loss (counted in full in the mod) and excess loss (discounted). NCCI sets it by state, and it is printed on the experience rating worksheet.
- Risk Management Alternatives: The Risk Management Matrix · Saylor Academy (open textbook)
- How to compute nonfatal incidence rates · U.S. Bureau of Labor Statistics
- CPCU 551 excerpt: the risk management process · The Institutes
Worked examples use illustrative numbers unless a source is named. This entry explains general concepts; it isn't legal, tax or coverage advice for a specific policy or contract.