Actuarial Software Efficiency & The Impact on The Implementation of IFRS 17

Recently, Alex Tsai, our Head of Greater China, spoke with Insurance Insight Summit in Beijing to take a deep dive into the critical question: Can the efficiency of actuarial software really have such a significant impact on the implementation of IFRS 17? The actuarial landscape is undergoing a paradigm shift, especially with the global adoption of IFRS 17, leading property insurance companies to explore the use of life insurance actuarial software for assessments. This piece explores the traditional boundaries of actuarial software, highlighting the shortcomings of predictive capabilities in traditional life insurance actuarial models.

With a focus on a "prediction model" example using R3S actuarial software, the blog also addresses the efficiency and impact of actuarial software under IFRS 17, particularly in the context of the upper-layer measurement model. Join us on this journey as we unravel the complexities and significance of actuarial software in the evolving landscape of insurance assessments.

Background

The actuarial model is the fundamental tool for conducting actuarial assessments. Traditionally, life insurance companies generally use specialized actuarial software for modeling, while property insurance companies mostly use statistical software or self-developed systems for model building.

However, with the rapid changes in accounting standards and regulatory rules, especially with the implementation of IFRS17 on a global scale, some property insurance companies are starting to use life insurance actuarial software for actuarial assessments. Some domestic property insurance companies are also exploring these related practices. It can be said that actuarial software is breaking through the traditional boundaries of life insurance use and gradually becoming a universal tool for various actuarial assessment tasks.

The shortcomings of traditional actuarial models in predictive capabilities are as follows:

In recent years, there has been an increasing demand for enhanced functionality and efficiency in the predictive aspects of actuarial models, driven by activities such as IFRS 17, solvency regulations, and valuation assessments. Traditional life insurance actuarial models often exhibit certain deficiencies in this regard, leading to significant pain points during their usage.

For instance, both solvency stress testing and quarterly solvency reports require solvency predictions. Despite the relatively short forecasting horizon for solvency predictions, regulatory provisions still allow the use of simplified methods such as the risk carrier factor approach under certain conditions. In the context of embedded value and new business value assessments, where the minimum capital prediction period is longer, the existing "Embedded Value Assessment Standards" do not impose restrictions, allowing the use of carrier methods for simplified calculations.

The permission to use these simplified methods is evidently associated with the limitations of traditional life insurance actuarial models in predictive functionality. This association leads to a situation where, for practical operability, a compromise may be necessary, sacrificing some precision and rationality in solvency predictions and value assessments.

Furthermore, the implementation of IFRS 17, which requires simultaneous prediction of future cash flows at the time of assessment, compels the industry to confront issues related to model predictions.

The two basic types of actuarial models.

Generally speaking, actuarial models possess two fundamental functions: point-in-time assessment and prediction. The emphasis on either point-in-time assessment or prediction during the initial design will result in significant differences in the model's predictive capabilities. From a functional perspective, life insurance actuarial models can be broadly classified into two basic types: "assessment models" and "prediction models."

Among them, "assessment models" primarily focus on point-in-time assessment functionality while also considering predictive capabilities. On the other hand, "prediction models" are built based on predictive functionality, with point-in-time assessment being automatically included as a special form of prediction.

Distinguishing between the two types of actuarial models

Point-in-time assessment and prediction are both essential functionalities for life insurance actuarial models. Why, then, is there a need to distinguish between two types of models? This is because, in most actuarial work, it is often necessary to simultaneously use multiple sets of different actuarial assumptions. This is manifested in two aspects:

1. The optimal estimate assumptions, adverse scenario assumptions, and comparison scenario assumptions in liability assessment.

2. The predictive assumptions and liability assessment assumptions in liability forecasting.

The technical root of distinguishing between the two types of models lies in how actuarial models simultaneously handle multiple sets of different actuarial assumptions.

In the case of "assessment models," when designing the model, the primary focus is on point-in-time assessment functionality. To meet the requirements of predictive functionality, a certain form of reset calculation is introduced in the model. This involves using the same set of calculation logic, calling different actuarial assumptions separately, and completing calculations under multiple sets of assumptions (can be understood as horizontal reset calculations between different assumptions). Additionally, in the predictions under each set of actuarial assumptions, the calculation logic from the assessment point is reused to complete calculations for each predicted time point (can be understood as vertical reset calculations along the timeline). For example, this can be achieved using features like Revaluation, Recalculation, Rebasing provided by actuarial software, or by writing code in the actuarial model to implement similar functionality.

On the other hand, in the case of "prediction models," when designing the model, the primary consideration is the implementation of predictive functionality. Point-in-time assessment is merely included as a special case. Therefore, the model architecture often differs significantly from that of "assessment models." For instance, hierarchical calculations may be used instead of reset calculations. This involves hierarchically setting up modules in the model to use different actuarial assumptions, achieving calculations under different assumptions in the corresponding hierarchical modules, and interacting the calculation results across different hierarchies.

A brief comparison of the two types of actuarial models

The "assessment model," without considering operational efficiency, is generally applicable to various actuarial assessment tasks and can even be considered to have a certain level of "universality." It excels particularly in point-in-time assessments. However, when it comes to implementing predictive functionality, the efficiency is often constrained due to the significant volume of reset/repeat calculations.

On the other hand, the "prediction model" generally has an advantage in implementing predictive functionality, but it may not be the optimal solution for point-in-time assessments. Moreover, the "prediction model" is often not as straightforward to implement the "universality" as the "assessment model." However, alternative methods can be employed to achieve a similar outcome, although this involves numerous technical details that won't be discussed here.

In summary, in actuarial assessment tasks, there is no absolute distinction between the two types of models in terms of superiority or inferiority; it depends on the specific analysis of the problem at hand.

An example of a "prediction model.

Due to the extensive discussions within the industry regarding traditional "assessment models" and their architectures, this article will focus on providing a simple demonstration of the "prediction model" and its architecture using R3S actuarial software as an example.

R3S actuarial software is a product of RNA Analytics. RNA Analytics is a software and consulting services provider with a history spanning several decades, headquartered in the UK and South Korea. The company is dedicated to leveraging technological innovation to assist the insurance industry in building an actuarial platform that spans regulatory compliance, risk management, and reporting systems. Currently, R3S has over 150 insurance institutional users across 49 countries and regions globally, including some of the leading insurance groups. In the domestic market, R3S also has a considerable number of clients. The R3S software features hierarchical calculations and modular modeling, making it relatively easy to implement the specialized architecture of a "prediction model."

Our model exhibits relatively minor differences in runtime duration between point-in-time assessment and prediction under the old standards (CGAAP) and solvency (CROSSII). Moreover, in various major actuarial application scenarios within the domestic context, it achieves high operational efficiency. As a result, there is no need to use a simplified carrier method for value assessment to predict the minimum capital for insurance risk.

The ability to achieve the aforementioned operational efficiency is primarily attributed to the functionalities of R3S, such as hierarchical calculations and modular modeling, making it relatively convenient to implement the architecture of a "prediction model."

In the modeling process for accounting reserves under the old standards, R3S users can place the same set of cash flow forecast modules (Programs or Projection processes) at different levels (Layers) within the model. They can then respectively read optimal estimate assumptions and adverse scenario assumptions to achieve both liability assessment and prediction.

In the solvency modeling process, R3S users can also, at different levels (Layers) within the model, use different cash flow forecast modules (Programs or Projection processes) to achieve solvency prediction functionality.

The efficiency of actuarial software under IFRS 17 and its impact on the implementation of projects: How much do we know?

The discussion above primarily focused on issues related to traditional actuarial models (cash flow models). However, the implementation of IFRS17 involves not only the cash flow models calculated on a policy-by-policy basis, known as the "bottom layer model," but also the upper layer model calculated on a group-of-contracts basis, i.e., the measurement model at the level of contract groups.

As the number of contract groups is often much smaller than the number of policies/model points, and the measurement at the contract group level is relatively simpler compared to the cash flow model, generally, there should be no efficiency issues with the upper layer model under IFRS17 (for example, as mentioned earlier, the upper layer measurement model of a domestic life insurance company using R3S takes only 1-2 seconds to run).

However, based on the current industry implementation experiences, the actual situation is far from ideal. Therefore, it is necessary to specifically address the issues related to the upper layer measurement model under IFRS17.

The following perspectives and materials are derived from the experiences and lessons learned from users who have already implemented IFRS 17 in the domestic market, with the hope of providing reference points for companies initiating projects this year.

In recent years, the implementation of IFRS 17 has been a major focus in the domestic insurance industry. Starting from the systematic launch by listed companies in 2023, to the ongoing implementation efforts by bank-affiliated and foreign-affiliated companies for the consolidation reporting requirements, the industry has witnessed a variety of solutions.

Whether led by the company's actuarial or finance department, and whether opting for integrated software (measurement engine + accounting engine) or completing measurements in existing actuarial software and then outputting results to sub-accounts, it is inevitable to achieve automated data flow between upstream and downstream systems. The pivotal role of measurement models/measurement platforms cannot be bypassed. This compels actuarial departments to step out of their traditional technical and operational domains, gradually merging with finance departments and even IT departments. The previous delineation of responsibilities between "you and me" becomes increasingly challenging, and "mutual interdependence" and "shared glory and disgrace, advance and retreat together" are becoming trends.

The pivotal role of the upper-layer measurement model/measurement platform under IFRS 17 is crucially manifested in the fact that the efficiency of the upper-layer measurement model/measurement platform will directly determine the overall implementation effectiveness of the IFRS17 project.

In essence, the efficiency of actuarial software is pivotal for successful IFRS 17 implementation, marking a shift where life insurance actuarial tools are embraced across the insurance landscape. The distinction between "assessment" and "prediction" models highlights the need for tailored approaches, illustrated by the R3S software's versatility. Despite expectations, challenges persist in the efficiency of the upper-layer measurement model under IFRS 17, emphasizing the necessity for collaboration among actuarial, finance, and IT departments to ensure a seamless transition and overall project success.