IAI CB3 key terms

A four-stage framework for managing risk: specify the problem, develop a solution, monitor results, and repeat as conditions change. It links professional judgement to real-world feedback loops.

The first control-cycle stage. It defines the risk situation, stakeholders, data, constraints, and the criteria a solution must satisfy before any modelling or product design begins.

The feedback stage of the control cycle. Actual experience is compared with assumptions, differences are analysed, and the model or strategy is revised so it stays relevant.

The standards, ethics, and conduct expected of an actuary, including integrity, competence, confidentiality, and communication of uncertainty. It governs how technical work is delivered and reported to clients and regulators.

The organised body of practitioners with qualifications, codes of conduct, disciplinary processes, and continuing education. Membership signals adherence to standards beyond purely technical skills.

A situation where a personal or commercial interest could improperly influence professional judgement. It must be disclosed and managed, often by declining the work or adding independent oversight.

Independent scrutiny of an actuary's work by another qualified person before it is relied upon. It checks assumptions, methods, and reasonableness, reducing error and supporting professional accountability.

The chance of an outcome differing from what is expected, including both upside and downside. In insurance it is usually framed as the possibility of loss or of liabilities exceeding assets.

Lack of certainty about future outcomes, which may be measurable (probability distributions) or unmeasurable (unknown unknowns). It is the raw material that risk management and modelling try to reduce.

The amount and type of risk an organisation is willing to accept in pursuing its objectives. It guides limits, capital targets, and product strategy, and must be approved at board level.

The practical limits within which risk appetite is expressed, such as maximum exposure per counterparty or a capital buffer. Tolerances are measurable and operational, unlike appetite statements.

A structured log of identified risks, each with an owner, likelihood, impact, controls, and mitigation status. It supports prioritisation and is reviewed regularly as part of governance.

Combining risks whose outcomes are not perfectly correlated so total variability falls below the sum of individual variabilities. It reduces required capital and underpins pooling in insurance.

A measure of how two variables move together, ranging from -1 to +1. Positive correlation weakens diversification benefits; assuming too low a correlation can understate aggregate risk.

Gathering many independent exposures so the average outcome becomes more predictable than any single one. It is the core mechanism by which insurers convert individual uncertainty into manageable aggregate cost.

A statistical result stating that as sample size grows, the observed average converges toward the true expected value. It justifies pooling and explains why larger portfolios are more stable.

A result that the sum or average of many independent variables tends toward a normal distribution regardless of the underlying distribution. It supports approximate aggregate risk modelling.

The probability-weighted average of all possible outcomes. It is the central estimate used in pricing and reserving, but it says nothing about the spread of outcomes around it.

The average squared deviation of outcomes from their mean, measuring spread. Its square root is the standard deviation, which is easier to interpret and often used in risk work.

The square root of variance, expressing dispersion in the same units as the data. It is widely used to describe volatility of claims, investment returns, and model outputs.

The standard deviation divided by the mean, giving relative variability. It allows comparison of risk across portfolios or variables with very different scales.

A symmetric bell-shaped distribution defined by its mean and standard deviation. It is convenient but often a poor fit for skewed insurance claims and heavy-tailed risks.

A measure of the asymmetry of a distribution. Positive skew means a long right tail, typical of large-loss claims, which normal-based models can dangerously understate.

The risk of extreme outcomes in the far ends of a distribution, beyond usual experience. It drives solvency capital and reinsurance needs and is poorly captured by averages alone.

The loss threshold that is not expected to be exceeded with a given probability over a defined period. It is simple to communicate but ignores the size of losses beyond the threshold.

The average loss conditional on losses exceeding a chosen percentile. Also called conditional tail expectation, it captures tail severity and is generally more conservative than value at risk.

A model that produces a single outcome from fixed inputs, with no random variation. It is simple and transparent but cannot show the range of possible results.

A model that incorporates random variation, generating a distribution of outcomes rather than one figure. It supports risk measurement but needs careful assumptions and can be computationally heavy.

A technique that repeatedly samples random values from assumed distributions and runs the model each time. The resulting set of outcomes approximates the full distribution of possible results.

The process of producing values that follow a specified probability distribution for use in simulation. Good generators are reproducible with a seed and statistically sound.

Evaluating results under a small number of specified, often extreme, situations rather than a full distribution. It is useful for exploring plausible stresses and communicating with non-specialists.

Assessing the impact of severe but plausible adverse conditions on a portfolio or firm. It reveals vulnerabilities that statistical models calibrated on normal experience may miss.

The risk that a model gives misleading results because of wrong assumptions, poor data, or implementation error. It is managed through validation, documentation, and limiting reliance on any single model.

Independent assessment that a model is fit for purpose, covering data, assumptions, methodology, and outputs. It includes back-testing and sensitivity checks before the model is relied upon.

Comparing model predictions with actual observed outcomes over a past period. Persistent discrepancies signal that assumptions or parameters need revision.

Testing how results change when one or more assumptions are varied. It identifies which inputs matter most and highlights where better data or judgement is needed.

The fitness of data for its intended use, judged by accuracy, completeness, consistency, and timeliness. Poor data undermines every later stage of analysis and must be documented and addressed.