No-Experience Risks And Zero Exposure

Why No Experience Gets the Collective Estimate

A credibility estimate balances risk-specific experience against a collective mean. If a risk has no observed exposure, it has supplied no experience to that balance. Its estimate therefore equals the collective complement: the mean rate for the relevant group of risks. This is not a penalty for having no claims, nor an assumption that the risk has a zero expected loss. It is the estimate supported by the information available.

The distinction matters in experience rating and classification. A risk with no exposure has no observed rate to average. By contrast, a risk with positive exposure and zero losses has an observed rate of zero for that period. Those records provide different information and must not be treated interchangeably.

Definition: In the Bühlmann-Straub model, let $\Theta$ represent a risk's underlying characteristics and let $X_{ij}$ be its observed rate in period $j$, based on exposure $w_{ij}$. Conditional on $\Theta$, assume $E[X_{ij}\mid\Theta]=\mu(\Theta)$ and $\operatorname{Var}(X_{ij}\mid\Theta)=\sigma^2(\Theta)/w_{ij}$, with observations conditionally independent across periods. Define the collective mean $m=E[\mu(\Theta)]$, the expected process variance $v=E[\sigma^2(\Theta)]$, and the variance of hypothetical means $a=\operatorname{Var}(\mu(\Theta))$. For positive total exposure $W_i=\sum_j w_{ij}$, the credibility estimate is $Z_i\bar X_i+(1-Z_i)m$, where $\bar X_i=\sum_jw_{ij}X_{ij}/W_i$, $Z_i=W_i/(W_i+K)$, and $K=v/a$, assuming $a>0$.

The weights $w_{ij}$ measure exposure in the chosen units. A period with $w_{ij}=0$ contributes neither exposure nor a rate observation to the risk's weighted mean. If every period has zero exposure, then $W_i=0$; the weighted mean formula has a zero denominator and is not defined. The zero-exposure case is handled directly: the estimate is $m$, not a calculated zero rate.

Why the Estimate Is the Collective Complement

For positive exposure, the credibility formula combines the risk's weighted observed mean with $m$. As total exposure decreases toward zero, $Z_i=W_i/(W_i+K)$ decreases toward zero, so the weight on the risk's experience vanishes. At exactly zero exposure, however, $\bar X_i$ does not exist. The formula's interpretation is that no risk-specific observation is available, leaving $m$ as the estimate.

\widehat{\mu}_i = \begin{cases} Z_i\bar X_i+(1-Z_i)m, & W_i>0,\\ m, & W_i=0. \end{cases}

A short prediction argument makes the zero-exposure result precise. With no observed data for the risk, any estimate based on its own observations must be a constant $c$: there are no observations to vary with. Under squared-error loss for predicting its underlying mean $\mu(\Theta)$, the best constant minimizes $E[(\mu(\Theta)-c)^2]$. Expanding around $m=E[\mu(\Theta)]$ gives

E[(\mu(\Theta)-c)^2] = \operatorname{Var}(\mu(\Theta))+(m-c)^2.

The variance term does not depend on $c$, and the squared term is minimized at $c=m$. Thus the collective mean is the best constant predictor of the risk's underlying mean when there is no risk-specific information. This argument assumes the target is the underlying mean and squared error is the criterion; it does not claim that the risk's true mean equals $m$. It says that $m$ is the estimate supported by the available information.

Worked Example: A New Risk With No Exposure

A hypothetical commercial class has collective mean $m=\$240$ per exposure unit, expected process variance $v=900$, and variance of hypothetical means $a=400$. A newly listed risk has zero exposure in every historical period. Estimate its rate and predict its expected losses at a future exposure of 3.5 units.

Method and assumptions. Apply the Bühlmann-Straub zero-exposure case. The class rates are expressed in dollars per exposure unit; the supplied model moments are $m=\$240$, $v=900$, and $a=400$. No parametric distribution is assumed; only the Bühlmann-Straub moment assumptions are used. Interest and mortality assumptions are not relevant to this rate estimate.

The structural constant is $K=v/a=900/400=2.25$. The risk has total exposure $W_i=0$, so its weighted observed mean $\bar X_i$ is undefined. Its credibility factor is $Z_i=0/(0+2.25)=0$, but the estimate must be interpreted using the zero-exposure case rather than multiplying zero by an undefined mean.

Therefore, $\widehat{\mu}_i=m=\$240$ per unit. At future exposure 3.5, the predicted expected losses are $3.5\times \$240=\$840$. Check: multiplying the collective rate by the stated future exposure gives a positive forecast of \$840, not zero. The common mistake avoided is treating “no historical exposure” as “zero historical losses.”

Finding the Collective Mean Without the No-Experience Risk

The collective mean may be supplied or estimated from the relevant portfolio. When it is estimated from observed risk rates, each rate is weighted by its exposure. A risk with zero exposure contributes no numerator and no denominator to an exposure-weighted portfolio mean. Including it as if it had a zero rate would introduce information that was never observed.

Worked Example: Calculating the Complement From Other Risks

In a hypothetical property class, risk A has exposure 5 units and an observed rate of \$110 per unit; risk B has exposure 3 units and a rate of \$150; and risk C has exposure 2 units and a rate of \$200. Risk R has no observed exposure. Estimate the collective mean and then estimate R's future expected loss at 4.2 exposure units.

Method and assumptions. Use the exposure-weighted portfolio mean as the collective estimate for this calculation. There is no assumed parametric distribution; the Bühlmann-Straub model is specified by its moments, and the rates and exposures are as stated. Interest and mortality assumptions do not apply.

First calculate each observed risk's exposure-weighted amount: A contributes $5\times \$110=\$550$, B contributes $3\times \$150=\$450$, and C contributes $2\times \$200=\$400$. The total is $\$550+\$450+\$400=\$1{,}400$, and the positive exposure total is $5+3+2=10$ units. Thus the collective mean is $\$1{,}400/10=\$140$ per unit.

Risk R contributes nothing to either total because its exposure is zero; it is not added as a zero-rate observation. With no risk-specific experience, R's estimate is the collective mean, $\$140$ per unit. Its future expected loss is $4.2\times \$140=\$588$. Check: the contributing rates lie between \$110 and \$200, and the calculated weighted mean of \$140 lies within that range. The common mistake avoided is dividing by a total exposure that incorrectly includes R.

Zero Exposure Is Not a Zero-Loss Observation

Suppose a risk has positive exposure but no losses in a period. Its observed loss rate for that period is zero, and the observation can affect its weighted mean and credibility estimate. The result depends on the exposure weight and model parameters. If the risk has zero exposure, there is no rate observation for that period at all. Even if an administrative file displays “0 losses,” that entry does not create a zero rate when the denominator is zero.

Record	Exposure	What it means for credibility
Positive exposure, zero losses	Greater than zero	An observed zero rate for that exposure
No exposure in the period	Zero	No rate observation and no contribution to the risk's weighted mean
No exposure in every period	Total exposure $W_i=0$	Estimate the risk mean by the collective mean $m$

Worked Example: Avoiding a False Zero Rate

A hypothetical liability portfolio contains two observed risks: risk A has exposure 4 units and rate \$180 per unit, while risk B has exposure 6 units and rate \$120 per unit. A newly entered risk Q has zero exposure and no claims recorded. Calculate the collective mean, Q's estimate, and expected loss at Q's future exposure of 2.5 units. Also quantify the effect of the mistake of assigning Q one artificial exposure unit at a zero rate when calculating the collective mean.

Method and assumptions. Use the exposure-weighted collective mean and the zero-exposure rule. Rates are dollars per exposure unit, and no parametric distribution is assumed beyond the Bühlmann-Straub moment framework. Interest and mortality assumptions are not relevant.

Risk A contributes $4\times \$180=\$720$, and risk B contributes $6\times \$120=\$720$. Total observed loss amount is $\$720+\$720=\$1{,}440$; total observed exposure is $4+6=10$. The collective mean is $\$1{,}440/10=\$144$ per unit.

Q has zero historical exposure, so its estimate is $\$144$ per unit. Its expected loss at 2.5 future units is $2.5\times \$144=\$360$.

If Q is incorrectly assigned one artificial exposure unit at a zero rate, the calculated collective mean becomes $\$1{,}440/(10+1)=\$130.909\ldots$ per unit. To the nearest cent, that incorrect value is \$130.91, rather than \$144.00. The error arises because the artificial record acts like a real, positive-exposure zero-loss observation. Check: Q's actual exposure is zero, so it must change neither the observed loss total nor the observed exposure total. The correct future expected loss remains \$360.

Practical Checks and Common Traps

For a risk with no experience, first confirm what “no experience” means in the data. Check that every included period truly has zero exposure, rather than positive exposure with zero reported losses. Then identify the appropriate collective mean for the risk's class and use it as the estimate. If forecasting aggregate expected loss at a future exposure $w$, multiply the estimated rate by that exposure: $w m$. The resulting value is an expected amount, not a revised rate.

Do not set the estimate to zero. No exposure means no information about the risk's rate; it does not imply a zero expected rate.
Do not calculate a weighted mean with a zero denominator. When $W_i=0$, $\bar X_i$ is undefined. The zero-exposure rule gives the estimate directly as $m$.
Do not assign artificial exposure. A zero exposure entry contributes nothing to exposure-weighted totals.
Do not confuse the rate with the forecast amount. The estimate is $m$ per unit; expected loss at future exposure $w$ is $wm$.

Key takeaway: Under Bühlmann-Straub credibility, a risk with total exposure $W_i=0$ has no risk-specific rate information. Its estimated underlying mean is the collective mean $m$, and a future expected loss is calculated by multiplying $m$ by the future exposure.

Check Your Understanding

Use the zero-exposure rule carefully; distinguish a missing rate from an observed zero rate.

A risk has no exposure in any historical period. The applicable collective mean is \$95 per unit. What is its estimated rate? (Answer: \$95 per unit)
A portfolio's observed risks contribute \$2,400 of exposure-weighted losses over 20 units. A new risk has zero exposure. Find the collective mean and the new risk's estimated rate. (Answer: \$120 per unit for both)
A risk has zero historical exposure, collective mean \$175 per unit, and future exposure 3.2 units. Find its predicted expected loss. (Answer: \$560.00)
Two risks have rates \$80 and \$140 per unit with exposures 4 and 6 units, respectively. A third risk has zero exposure. Calculate the exposure-weighted collective mean, excluding the zero-exposure risk. (Answer: \$116 per unit)
A portfolio has exposure-weighted losses of \$3,600 over 30 units. A no-experience risk is mistakenly assigned two artificial exposure units at a zero rate. What is the incorrect collective mean, to the nearest cent, and what is the correct mean? (Answer: Incorrect \$112.50 per unit; correct \$120.00 per unit)

What You'll Learn