1. Using machine learning for reserving
Why should you consider using machine learning for reserving?
Overview
One of the most asked questions/challenges to the working party is why should reserving teams use machine learning (‘ML’) methods, when traditional methods coupled with actuarial judgement often generate reasonable (or even better at this stage) results. We see broadly 3 key areas of benefit from using ML methods in the reserving context:
- Insight
- Efficiency
- Flexibility.
Before we delve into the details, it is worth clarifying that we do not necessarily expect ML methods to be used to replace any existing reserving methods at least in the short term, and certainly not as a replacement for actuarial judgement. Instead, we see ML models as extremely useful extensions to the actuarial toolkit which could bring great benefits brought to reserving in today’s rapidly changing and data-rich world.
Insight
ML methods could be more effective in detecting trends compared to traditional methods, or even human judgements in some circumstances. As a result, ML methods could provide better or faster insight to the changing world around us, highlight the likely effect of these changes, and allow actuaries to adjust reserves to better prepare for the challenges inherent in the unknown. Expanding on this:
- ML methods may be more effective in detecting trends present in data compared to traditional methods - for example, it may detect calendar or accident year trends
- Experience from Pricing tells us that such additional information can often be explicitly incorporated into machine learning methods and doing so gives better results - the same may be true for reserving
- Additional insights can add value to insurers beyond setting reserves
- ML brings the potential to get earlier and more accurate indications of ultimate costs which will bring benefit to the Pricing and Underwriting and Claim functions
Efficiency
Using ML and related data science concepts appropriately could result in turning around reserving tasks in a fraction of the time that more traditional approaches would take. This benefit itself would enable reserving to be performed at more granular levels, more frequently, opening doors to new ways reserving may be operated and utilised to generate additional business value for insurers. Additionally, even if a company choses not to use ML models, there are potential efficiency gains from ML data handling principles that can improve the process.
In more detail:
- ML brings the potential to automate aspects of the Reserving function’s work, thereby freeing up more time for actuaries to apply their valuable judgment to risks outside of those seen in the past data which is an area that machine learning methods are not so good at dealing with
- ML may detect trends earlier and more accurately than many traditional methods, which will improve the efficiency of the reserving process
- You don’t even have to go as far as developing a full ML model right away to see the benefits - the pre-work that goes into developing these models can bring a lot of benefits to your team
- Clean-up / automate data processes
- Collecting and preparing more granular data may lead to fewer errors, more insights from the data, and put you in a better position to answer questions from stakeholders.
Flexibility
The predictive power of ML methods in data-rich environments, coupled with the increased level of automation compared to traditional methods (which often require higher level of human intervention) help to ‘future proof’ reserving processes from new critical information, changing risk environment, or even reporting changes.
- Many traditional reserving methods, including the Chain Ladder, make the assumption that development patterns observed in the past will continue to be relevant in the future. However this is often not the case (especially in recent times - wars, pandemics, inflation, climate change).
- ML models may be more agile at adapting to new experience and providing insights to actuaries to help them assess valid extrapolations for reserving
- The more granular data collection and management required by ML models will in turn allow actuaries to address a wider range of questions.
Individual responses
Finally we’ve included some individual comments on the advantages of using ML, including some from other material.
John
The predictive power of ML methods in data-rich environments is proven in other contexts. Even applying the methods to triangular data, which is easy to set up and covered in our blog, brings something to the party, for example the detection of calendar or accident period trends. Some of the tools that actuaries will acquire incidentally from trying out ML methods will also improve the reserving process for example the dashboard functionality in R Shiny, or the capability to automate some of the data preparation steps using code.
The world is increasingly data-rich. In other areas in finance and insurance, machine learning has been able to provide a deeper understanding of the underlying trends, and consequently provide more accurate predictions, in an automated way. The hypothesis here is reserving is no different - that there must be that same potential.
Jacky
The world is changing rapidly. Many traditional reserving methods, including the Chain Ladder, make the assumption that development patterns observed in the past will continue to be relevant in the future. In recent times, it has been demonstrated that this may no longer by the case.
Isabelle
With COVID still fresh in our minds and the spectre of climate change ever-present in the background, it is crucial that we invest in flexible processes that can better capture these changes in our environment, and predict how they might affect our future. Even changes in business mix made necessary by alterations in the wider environment could invalidate the assumptions inherent in these methods, in addition to introducing a further level of difficulty to the actuarial judgement required in setting an appropriate reserve. While adjustments can be made to traditional methods by explicitly altering assumptions, it can be difficult to detect whether or not these alterations might be required. ML models can give us insight to the changing world around us, highlight the likely effect of these changes, and allow us to adjust reserves to better prepare for the challenges inherent in the unknown.
Grainne
It’s true that ML offers the potential to get additional insights into the data and this is a good reason for using it. But even without that (noting that reserving actuaries are generally good at their jobs and understand their data very well), the automated side to ML has the potential to generate a more robust and efficient process. Using ML and related data science concepts could result in turning around an updated valuation estimate, together with data processing and checking, and subsequent management information in a fraction of the time that a more traditional approach would take. This is beneficial even before considering any potential improvements in accuracy.
Tom
Further to the above, I would note that ML is not Science Fiction; a number of providers in the London Market already use it to perform their reserving (one company in particular performs an ML based exercise, which takes a couple of hours, before embarking on a traditional reserving exercise to validate that number which takes several weeks/months).
This is certainly the direction things are moving in, with some airtime being given to ML but it not yet being fully trusted; over time we hope to see the weighting/reliance placed on ML methods gradually increasing as they become more familiar. I would compare the benefits to those of a stochastic model vs a deterministic one with specific scenarios envisaged; just like how the stochastic engine might shine light on that one niche (but very damaging) scenario that no one would have conceived of, so too can ML methods give you an “out of the box” result that causes you to challenge your existing thinking. Finally, I have found learning about ML techniques to be hugely enjoyable and intellectually nourishing; for an industry that is often accused of being slow to change, I think it’s very exciting to be on the cutting edge.