Exposure Data Quality: Units And Missing Values

Why Exposure Data Quality Matters

Exposure measures the amount of insured risk over a defined basis, such as vehicle-years, payroll, or insured value. The earlier material on exposure bases explains that the appropriate measure depends on the coverage and rating purpose. Before using exposure records, an analyst must also establish that each value is expressed in the expected unit and that the value actually means what the data record suggests.

A number alone is not enough. The value 500 might mean 500 dollars, 500 thousand dollars, or 500 vehicle-months. A blank might mean “not reported,” while zero might mean “the measured exposure was confirmed to be none.” Treating these cases alike can produce invalid totals and misleading rates. This tutorial presents checks for exposure units, missing values, and impossible amounts, aligned with the CAS Exam 5 (Basic Ratemaking and Estimated Claim Liabilities) syllabus.

Definition: A valid exposure value is a value with a known meaning, a documented unit, and a result that satisfies the applicable business rules for that field. A zero exposure is a valid numeric value only when the data or supporting information confirms that the measured exposure is zero. A missing exposure has no usable numeric value; it is unknown, not zero.

Standardize Units Before Comparing Values

A unit check begins with the definition of the field and the unit attached to each record or source. Select one canonical unit for the field, then convert each interpretable value to that unit. For example, a vehicle-month value can be converted to vehicle-years, and payroll reported in thousands of dollars can be converted to dollars.

Unit conversion rule: If a reported value $r_j$ uses a known unit and $c_j$ is the number of canonical units in one reported unit, its standardized value is $e_j=c_jr_j$. This conversion is valid only when the reported unit is known and the conversion factor is appropriate for the field. Do not infer a unit solely from the size of a number.

The reason for standardizing first is straightforward. Values can be added or compared as exposure amounts only when they represent the same measure in the same unit. If one record contains dollars and another contains thousands of dollars, their raw numeric sum is not a meaningful dollar total. The conversion must be applied record by record before aggregation:

\text{standardized total}=\sum_j e_j=\sum_j c_jr_j

This expression applies to records with known, valid values. A missing value has no $r_j$ to convert. It must not be inserted into the sum as though it were a confirmed zero.

Worked Example: Converting Vehicle-Months and Separating a Blank

A hypothetical insurer stores vehicle exposure in vehicle-months in one feed, but its reporting field is vehicle-years. The four records are 12 vehicle-months, 6 vehicle-months, an explicitly confirmed zero vehicle-months, and a blank. Convert what can be converted and state what can be concluded about the total.

Method. Standardize the known values to vehicle-years. By definition, 12 vehicle-months make one vehicle-year, so divide each known vehicle-month amount by 12.

Record 1: $12\div12=1.000$ vehicle-year.

Record 2: $6\div12=0.500$ vehicle-year.

Record 3: $0\div12=0.000$ vehicle-years. This is a confirmed zero, not a missing observation.

Record 4: the value is blank, so there is no numeric amount to convert. Its exposure is unknown.

The subtotal of the known, converted values is $1.000+0.500+0.000=1.500$ vehicle-years. This is a known-value subtotal, not a complete total for all four records. The blank record requires investigation or an explicit missing-data treatment before a complete total can be stated.

Check and common mistake. Converting back gives $1.000\times12=12$, $0.500\times12=6$, and $0.000\times12=0$ vehicle-months, matching the known inputs. A common mistake is to write “the total is 1.500 vehicle-years” as if all records were complete. The correction is to report 1.500 as the subtotal of known values and identify the remaining record as missing.

Missing Values Are Not Zeros

A field can be missing for many reasons: it was not collected, a source failed to transmit it, or a value was recorded using a code that has not been interpreted. Common representations include a blank cell, a null value, or a special code such as “NA.” The meaning of a special code must be confirmed from the data specification; it should not be treated as a numeric exposure merely because it appears in a numeric field.

A zero, in contrast, is a reported amount that can be meaningful. For example, a payroll exposure of zero might be supported by a verified statement that no payroll occurred in the measured period. Whether zero is plausible depends on the field and the record’s context. An explicit zero can still merit a consistency check, but it is not automatically missing.

Confirmed zero: retain as zero and preserve evidence that the value was intentionally recorded.
Blank or null: record as missing; do not silently substitute zero.
Unknown code: resolve its meaning from documentation or the data owner before interpreting it.
Unresolved value: keep it flagged and separate from the valid numeric values until investigated.

If the measure is nonnegative, adding only valid observed values can produce a subtotal, but not a complete exposure total when records are missing. Replacing missing values with zero would assert information the data do not provide. It can also make a group appear to have less exposure than it actually does.

Worked Example: Payroll in Dollars and Thousands of Dollars

A hypothetical commercial policy file contains three interpretable payroll entries and one flagged amount. Record A reports $480,000, Record B reports 525 in units of thousands of dollars, Record C reports an explicitly verified $0, and Record D reports $-\$1{,}200$. The reporting field is dollars. Standardize the values and decide what can be reported.

Method. Convert each known entry to dollars, then apply the sign check. One thousand dollars equals $1,000, so the conversion factor for Record B is 1,000 dollars per reported unit.

Record A is already in dollars: $\$480{,}000$.

Record B converts as $525\times\$1{,}000=\$525{,}000$.

Record C is an explicitly verified zero: $\$0$.

Record D is negative payroll. Because payroll exposure cannot be negative under the field definition, this amount fails the validity check. It is not a valid exposure to include in a total; it should be flagged for investigation.

The subtotal for the valid records is $\$480{,}000+\$525{,}000+\$0=\$1{,}005{,}000$. This is a valid-record subtotal, but a complete file total cannot be certified while Record D remains unresolved.

Check and common mistake. Record B converts back to $525{,}000\div1{,}000=525$ thousands of dollars, confirming the conversion. Adding the raw entries as if all were dollars would give $\$480{,}000+\$525+\$0=\$480{,}525$ before even considering the invalid record. That result is wrong because the units differ; convert first, and do not use the negative value as valid exposure.

Identify Impossible Amounts Using Field Rules

An impossible amount is one that violates the definition or an applicable business rule for its field. Some checks are general: exposure amounts ordinarily cannot be negative. Other checks depend on the field and the data specification. A maximum insured value, for example, should be based on an established coverage or system constraint, not an analyst’s intuition about what “looks large.”

A value outside an expected range is a reason to investigate, not proof of the correct replacement value. A suspicious amount could reflect a unit error, a data-entry error, a special coding convention, or an unusual but valid record. Preserve the original value and record the reason for the flag; do not silently alter it.

Worked Example: Validating Insured Values in Thousands

A hypothetical property file records building insured values in thousands of dollars. For this field, the documented rules require a value between 0 and 8,000 inclusive. The records are 1,250, 0 with an explicit “no building exposure” indicator, 9,600, $-15$, and a blank. Standardize the interpretable values and classify each record.

Method. Use the documented conversion factor of $1,000 per reported unit, then apply the field’s range rule. In dollars, the allowable range is $0 through $8{,}000\times\$1{,}000=\$8{,}000{,}000$, inclusive.

Record 1: $1{,}250\times\$1{,}000=\$1{,}250{,}000$. This is within the permitted range, so it is valid.

Record 2: $0\times\$1{,}000=\$0$. The zero is explicitly supported and is within the permitted range, so it is a valid zero for this field.

Record 3: $9{,}600\times\$1{,}000=\$9{,}600{,}000$. This exceeds the $8,000,000 maximum by $\$9{,}600{,}000-\$8{,}000{,}000=\$1{,}600{,}000$, so it must be flagged.

Record 4: $-15\times\$1{,}000=-\$15{,}000$. This is below the permitted minimum of zero, so it must be flagged.

Record 5 is blank and therefore missing. It has no numeric amount to convert or test against the range.

The valid-record subtotal is $\$1{,}250{,}000+\$0=\$1{,}250{,}000$. It is not a complete total for all five records: one is missing and two fail the documented range check.

Check and common mistake. Converting the valid positive entry back gives $\$1{,}250{,}000\div\$1{,}000=1{,}250$, matching the source value. Treating every numeric entry as valid would incorrectly add amounts that violate the rules. Conversely, the zero is not invalid merely because it is zero: its explicit meaning and the field rules support it.

A Practical Validation Sequence

A consistent validation sequence helps separate unit problems, missingness, and impossible amounts rather than allowing one issue to hide another.

Define the field.
State what the exposure measures, its intended basis, and the permitted values or range. Use the applicable field specification rather than assumptions based on appearance.

Identify units and codes.
Confirm each source’s unit and the meaning of blanks, nulls, and special codes. If the unit or code is undocumented, mark the record unresolved rather than guessing.

Classify missing and zero values.
Preserve a confirmed zero as a numeric amount. Keep missing or unknown values separate and retain the reason they are unavailable when known.

Convert and test.
Convert interpretable amounts to the canonical unit, then apply sign, range, and other field-specific checks. Keep flagged values distinct from valid values.

Report what is supported.
Label a sum of valid records as a subtotal if any records are missing or unresolved. State the counts or status of missing and flagged records so the subtotal is not mistaken for a complete total.

Common Traps and Why They Matter

The most consequential mistakes are often simple substitutions: treating all numeric entries as if they used the same unit, converting a blank to zero, or assuming that a value is valid because it can be added. These mistakes can change exposure totals and any later calculation that uses those totals.

Trap	Why it fails	Correction
Adding values from different units	The numeric sum does not represent a common measure.	Document the units and convert every valid value to one canonical unit first.
Replacing blanks with zero	It asserts that an unknown amount is known to be none.	Keep missing values separate; report a subtotal when completeness is unresolved.
Rejecting every zero	A zero can be a valid, verified exposure amount.	Check the field meaning and supporting indicator before classifying it.
Accepting a plausible-looking amount	A value can be in the wrong unit or outside documented limits.	Apply field-specific rules and investigate flagged values without silently changing them.

The central discipline is to preserve the distinction between what is known, what is missing, and what has failed a validity check. This makes later exposure summaries interpretable: users can see whether a reported amount represents all records or only the valid values currently available.

Check Your Understanding

For each problem, distinguish confirmed values from missing or flagged amounts, and state only the total supported by the information given.

A hypothetical file records 18 vehicle-months for one vehicle. Convert this to vehicle-years. (Answer: 1.5 vehicle-years.)
A payroll field is in dollars. A valid record is reported as 720 in thousands of dollars, and a second record is a confirmed zero dollars. What is the valid subtotal in dollars? (Answer: $720,000.)
Three vehicle-year records are 0 with an explicit confirmation, blank, and 4.5. What is the subtotal of known valid exposure, and is it a complete total? (Answer: 4.5 vehicle-years; no, one record is missing.)
A building-value field is recorded in thousands of dollars, with a documented allowed range from 0 to 2,000 inclusive. Values are 940, 0 with an explicit “no building exposure” indicator, and 2,100. Which values are valid, and what is the valid-record subtotal in dollars? (Answer: 940 and 0 are valid; 2,100 is flagged for exceeding the maximum; the valid subtotal is $940,000.)
A hypothetical exposure file contains 300 in thousands of dollars, a blank, and $-20$ in thousands of dollars. The field is nonnegative payroll exposure. State the valid-record subtotal and the unresolved issues. (Answer: $300,000 valid-record subtotal; one record is missing and one negative record is flagged, so the complete total is unresolved.)

What You'll Learn