Author: Guy Shepherd
Although financial modelling has been an essential and critical part of insurance operations for at least 25 years, the topic of assumption management continues to generate a significant level of debate between actuaries, finance, risk management, and IT. And despite the importance of assumption management, many organisations still struggle to find a robust yet flexible solution to the problem. As the number of assumption sets continues to grow in order to support a greater variety of modelling tasks, and with an increased focus on transparency, auditability & control, and with less working days to deliver results, the need for a good assumption management framework is back in focus again.
The fundamental question typically goes something like this:
“Has the proper team used the correct assumption sets, in the right version of the approved model, with the correct source data, to produce the right results for the required purpose on the right date?“
While this sounds like an obvious and relatively simple question to answer, it still presents a headache for many insurers. The larger and more complex the insurer, the more data sets, assumption sets, and models there are in use across more teams to produce a broader range of results, and so the trickier it is to answer and evidence the answer to that question.
The following list covers what might reasonably be considered as the 5 essential features of a robust assumption management framework, although there are interdependencies between many of them. Addressing these requirements should go most of the way to answering the fundamental question above.
1. Lineage & Traceability
The fundamental requirement here is to understand the ultimate source of an assumption or assumption set. – i.e. where have they come from and is that a trusted & expected source? Demonstrating the provenance of assumptions can be tricky, especially as some assumptions are part of the woodwork and are handed down from one reporting period to the next. Assumptions often come from disparate sources within the business and beyond, and it’s typical for assumptions to arrive by email or even word of mouth. Sometimes, it may not be possible to demonstrate lineage or look-through for all assumptions back to an empirical source, but that doesn’t mean the journey of assumptions through models shouldn’t be well controlled.
2. Transparency & Auditability
Promoting visibility of assumption sets across departments and their various uses is fundamental in creating transparency & confidence in an organisation. Assumptions should stand up to scrutiny across a business, to the extent that it should be possible to explain how different assumption sets are used for different purposes across the wider business community, and perhaps more fundamentally why differences exist between the different sets. Many businesses struggle with many versions of the truth when it comes to assumptions, simply because the availability and purpose of existing assumption sets is unclear. A good assumption management framework should also make it clear which versions of assumptions should and have been used for different purposes. Where assumptions have been changed for good business reasons, it’s also essential that those reasons are captured to help avoid confusion.
The objective here is to ensure that there is a consistent view and understanding of assumptions across all stakeholders in the actuarial and wider business community – e.g. risk, finance, customer services, etc.
Assumptions should have consistent definitions and naming conventions across different use cases to avoid misunderstanding and misuse. Even relatively modest differences of opinion or understanding can have a significant bearing on how assumptions are defined and used. For example, the definition of what constitutes a new business date might vary between actuarial & finance – actuarial may consider it to be the date the policy appeared on a policy administration system, whereas finance might consider it when the first premium is received.
One of the best ways to avoid issues with consistency is develop a business wide glossary or lexicon of terms, which is easily accessible to all and which specifically defines specific terms, and covers any potential conflicts to ensure the correct assumption definition is used.
4. Accountability & Approvals
Having mapped out the lineage of each assumption & assumption set as outlined above, it’s also important to understand who is accountable for ensuring the assumptions are correct as they flow through the business. Larger and more complex businesses will have any number of data stewards responsible for different stages of the assumption management process. It’s imperative to ensure that any stakeholders who might be considered as the ultimate owner / creator of any assumptions are made aware of their responsibilities. While this sounds an obvious step, it is often overlooked. Furthermore, obtaining a formal sign-off from each assumption owner & steward is a perfectly reasonable requirement.
Manual assumptions, which are typical in many business processes present a specific set of challenges. Any good assumption management framework needs to cope with the reality of discrete / manual assumptions, and they should offer the same level of control & transparency as any other type of assumption. Manual assumptions typically only come from a restricted group of trusted senior staff, but having a secondary or even tertiary approval step is also an essential requirement of a robust framework.
Similarly manual overrides, which often happen at the last minute for fully justified business reasons need to capture the what, the why, the who, and the when of the override, and also require 4 or 6 eyes approval.
5. Evidence & Reporting
The final feature of a good assumption management framework, which is often overlooked or only considered as an afterthought, is the ability to evidence the flow of assumptions from sources through to use, which a complete history of touchpoints, changes, and approvals. The best assumption management frameworks provide a comprehensive report of all the assumptions used in a given business process, together with a complete chain of custody starting with the ultimate source and capturing any changes & overrides. Such a report goes a long way in satisfying the needs of risk management, internal / external audit, and other downstream business customers, and answering the fundamental question posed at the start of this article. Attempting to produce such a report without having implemented the other features listed above is a real challenge and can be incredibly time consuming.
So, What’s the Solution?
Assumption management solutions generally fall into 2 broad categories:
Decentralised: assumptions are held & managed at a functional, departmental or team level.
Centralised: all assumptions are held & managed by a specific team in a central repository.
Most insurers have historically used a largely decentralised approach to assumption management. Departments and teams have had responsibility for sourcing, manipulating, storing, and using their own assumption sets appropriate to their specific business activities. These assumptions are often held in spreadsheets and are stored in departmental folders, which are typically duplicated for each reporting purpose and cycle.
Meeting even a subset of the requirements listed above for decentralised assumptions can be a real challenge. The governance of decentralised assumptions is especially challenging, given the ease by which files can be changed without any robust approval mechanism. Likewise, evidencing the journey of assumptions from source through models into results is a real challenge.
A centralised approach to assumptions management will typically see assumptions stored and managed by a central team in a single database or data warehouse solution. This can present a challenge with legacy / first generation modelling platforms, which are unable to integrate directly with these data stores and rely instead on assumptions being provided as files or in proprietary formats, which undermines some of the transparency & consistency considerations of a good framework.
But simply storing assumptions in a database is not the complete solution. In particular, specific thought must be given to the controls placed around the journey of each assumption into the central assumptions database. Furthermore, careful consideration must be given to the management of assumption changes and the version control of assumption sets once assumptions reach the database. While database technologies can certainly help provide a robust solution, it’s unlikely to be provided on a plate.
Assumptions Management with Mo.net Solution
In a future article I will explore how the various development and operational components of the Mo.net platform have been explicitly designed to support a robust assumptions management framework, in both decentralised and centralised environments.