Data Quality Fundamentals

The Dimensions of Data Quality

At a high level, data quality is about ensuring that data is fit for its intended purpose. While this sounds simple, it encompasses several distinct dimensions. To begin, let's get a general business perspective on what constitutes "good" data.

The video provides a helpful starting point, highlighting concepts that are universally important. In actuarial practice, these ideas are consolidated and formalised. While terms like "consistency" and "uniqueness" are vital, they are often considered aspects of the three core pillars of data quality that you must know for your exam:

1. Accuracy
2. Completeness
3. Appropriateness

Let's explore each of these in the specific context of actuarial work.

The Three Pillars of Actuarial Data Quality

The Casualty Actuarial Society (CAS) and other professional bodies have published extensive guidance on data quality. We will draw on these to build a robust understanding of each pillar.

Drawing from this paper and broader actuarial practice, let's refine our definitions.

1. Accuracy

Accuracy answers the question: Does the data correctly reflect the real-world facts it is intended to represent?

A key distinction the CAS paper makes is between validity and accuracy.

• Validity: The data conforms to the required format. For example, a "Date of Birth" field containing "30/02/1990" is invalid. A field containing "01/05/1985" is valid.
• Accuracy: The valid data is also correct. The date "01/05/1985" is valid, but it is only accurate if that is the person's actual date of birth.

An insurer might have a policy record where the gender is coded as 'M'. This is a valid entry, but if the policyholder is female, the data is inaccurate. Such an error could have a material impact on pricing or reserving for a life insurance product.

The CAS paper also introduces a practical hierarchy of accuracy that is very useful in the real world:

• Absolute Accuracy: The data is 100% correct. This is the ideal but is rarely achievable in large, complex datasets.
• Effective Accuracy: The data contains some imperfections, but they are not material to the analysis at hand. For example, if a small number of postcodes are incorrect, it might not materially affect a nationwide reserving exercise, but it would render an analysis of risk by geographic area useless.
• Relative Accuracy: The data is defined or coded incorrectly, but it is done so consistently over time. For instance, a company might consistently include a certain type of expense in its claims data that, by definition, should be excluded. An actuary can still use this data for trend analysis, provided they are aware of and can adjust for this inconsistency.

2. Completeness

Completeness asks: Is all the necessary data available?

This extends beyond simply checking for blank fields. As the CAS paper outlines, completeness has three essential elements:

1. All necessary fields are present for each record. For example, a motor insurance record without a vehicle type is incomplete.
2. All necessary records are present, and only once. This means no duplicate policy records and no missing ones. Reconciling premium income in the policy data to the finance department's general ledger is a common check for this.
3. All records have been processed correctly through the necessary systems.

The challenge of ensuring completeness across different data sources is a major modern data-governance problem, especially with the rise of complex analytical models.

3. Appropriateness

Appropriateness, or suitability, is the most judgmental pillar. It asks: Is this data suitable for the specific actuarial task I am performing?

Data can be perfectly accurate and complete, yet entirely inappropriate. For example, using a company's claims data from 20 years ago might be inappropriate for pricing a new product today because the economic environment, product features, and policyholder behaviour have all changed. The old data is no longer representative of the risk being modelled.

The CAS paper highlights key questions to assess appropriateness:

• Is the data the information actually needed for the analysis?
• Is it sufficiently homogeneous (i.e., does it represent a consistent group or phenomenon)?
• Is it consistent with the overall purpose of the study?

A common check for appropriateness is to assess its reasonableness. Does the data, when aggregated, make sense? For example, if an analysis of new business shows the average age of annuitants has suddenly dropped from 65 to 35, this is unreasonable. It doesn't mean the individual data points are inaccurate, but it strongly suggests the dataset is inappropriate, perhaps due to a data extract error that has pulled in a different product type.

Data Quality in Professional Standards and Practice

These principles are not just best practice; they are embedded in the professional standards that govern actuarial work. The UK's Financial Reporting Council (FRC) publishes Technical Actuarial Standards (TASs) that actuaries must follow.

As TAS 100 makes clear, an actuary's responsibility goes beyond just using the data they are given. They must actively:

• Ensure effective checks and controls are applied.
• Identify material bias (e.g., if a dataset over-represents a certain demographic, making it not representative of the whole portfolio).
• Take steps to improve the data or, if that's not possible, assess and communicate the limitations.

To see how a company might implement these requirements in practice, look at the following example of a data audit framework.

This template translates the principles of TAS 100 into a concrete, auditable process. The "Expected Control" column lists specific requirements for a data policy, such as defining data ownership and setting quality standards. The "Assessment Approach" column lists the evidence an auditor would look for, such as written policies and documented procedures. This demonstrates how data quality is managed as a formal risk control within a financial institution.

Conclusion

Ensuring data quality is a foundational habit of any effective actuary. It is the first and most critical step in managing the "Input Risk" we identified in our previous lesson. We've seen that high-quality data must be accurate, complete, and appropriate for its intended use. These are not just abstract ideals but core professional obligations, enshrined in standards like TAS 100 and implemented through rigorous internal controls and audit processes.

Understanding these dimensions is the first step. In our next lesson, we will build directly on this foundation to explore the practical "how-to" of data management. We will delve into the specific processes actuaries use for data verification, cleaning, and making adjustments to deal with the inevitable imperfections found in real-world data.

Key Takeaways & Cheatsheet

Here is a framework to help you remember the key dimensions of data quality and the questions you should ask for each.