Dangers and Blind turns of using statistical models only:
The insurance industry has historically been a mix of human underwriting/”gut feel” vs. actuarial/statistical models. If you ask experienced or retired professionals about the evolution of the industry, they might say – it was more of the former and less of the latter. This balance has changed over time to a high reliance on the statistical and actuarial outputs for making decisions. Senior leadership and investors have more faith in deploying risk based capital to decisions that the model approved/supports vs. gut feel…but why? is this appropriate?
Historic success of the traditional system
I am not preaching or coming from the position that the insurance system cannot be improved and that if I were to start it from scratch today, I would do it the same way. Actually I am in the opposite corner of the argument, but the industry has experienced success in transferring risk, funding liabilities, generating a return for investors, and overall positive impact on society for hundreds of years.
These historic successes of the insurance system were built using a blend of underwriting (human) decision making and actuarial (statistical models).
Allowing the industry to take on bigger risks and insure more parts and different risks of society.
Risk takers and Investors
However, the gap between economic and insured losses is growing – partly due to the new nature of risks and lack of data – also partly due to the industry shift and reliance on models to guide decisions. Less “gut feel” and Art of underwriting more statistical model driven decision making. Resulting in analysis paralysis.
This has resulted in an easy out for forecasters (Senior leaders and underwriters). When questioned about their failure to appropriately forecast an outcome, at least they are defensible decisions… “The model told me so!!!”
If that’s the case – investors should be asking – why are we the investors paying for the “talent” when the model is doing all the work?
Relevance of Insurance
This is a question that is occurring as investors and the industry reviews it’s relevance.
The over reliance on models has some leaders and forecasters paralyzed in their ability to take sizable or appropriate risks which is causing a wider gap between insured and economic losses.
This gap in relevance (measured by economic vs. insured losses), lack of investment income, as well as the industry return pressures have allowed for industry outsiders to try and solve some of the problems associated with the insurance system.
With the introduction of new Fintech/Insurtech firms in the insurance industry, more and more firms are focusing on the inefficiencies of the industry.
Many firms are focusing on the issue of having an actuary and human review the data with a slow manual decision. Many are exclaiming that there is plenty of room for human error, the expensive nature of humans, and lack of scale that human work can produce (actuaries don’t work 24/7 for those of you who are unaware of there limitations), and on & on.
The Fintech/Insurtech approach bringing new forecasting tools associated with machine learning and statistical models will likely help with relevance of the industry, but what risks or issues do they present:
- What are the dangers and blind turns of using statistical models only?
- Is the industry and it’s investors willing to accept that risk to lower the expense ratio?
Data relationship issues
One of the risks associated with statistical models is that there are false positives or negatives in data that obscure the relationship between factors. This was historically resolved by an underwriter and actuary sitting in a room discussing the data – and either smoothing the data, curves, or assumptions due to the anomaly or false positive or negative in the data.
The ability to scale a statistical model or algorithm to forecast insurance liabilities is great and understandable exciting. The ability to forecast 24/7, 365, etc. plus at a low cost with no human related expenses is easy to see it’s attraction.
However, a statistical model or algorithm alone may not catch the data issues that the human and actuary were historically reviewing together.
- What happens then? When will data errors be caught?
- How big is the false positive or negative?
- Is the error is loss pick or forecasting worth the expense savings?
- We have seen high frequency trading and algorithmic tools fail before, will a human be able to reverse the decision? And at what cost?
- Will this outcome above be more volatility then investors are accustomed to or less?
Historical data not exact prediction of the future
Anyone who has worked with data before knows historical data is never an exact predictor of the future. All forecasters understand this and make assumptions to help them forecast – a statistical or algorithmic model only approach must take into account this phenomenon. Therefore, the model is only as good as the assumptions made or forecasters on the team developing the tool.
Positives of statistical and economic capital models
There are many positives of statistical, algorithmic, and economic capital models – a few include:
- Appropriate funding of insurance liabilities and asset matching strategies
- Lower cost of capital due to Lower volatility
- Ultimately resulting in lower cost to consumers
Be wary of investing in insurance companies that say the statistical models and tools are not relevant to their success. Also, be wary of those statistical modelers or salesman who state human influence is not important to results. These activities are both essential and intimately related to the success of an insurance company.
Now certain improvements and many efficiency could be introduced to most companies. But what distinguishes good insurance company underwriters and management from the bad ones is the blend between:
- the Art,
- Statistical modeling
Ultimately you will see it in the Financial results.
The companies that find the right blend of Insurtech and traditional analysis will see the best results and should also see the greatest reward from investors.
- How Insurance could become the Innovation industry
- What insurance industry can learn from baseball?
- Machine learning algorithm vs Actuarial science
- Programming a computer for evaluating Insurance risk
- Be cautious about insurers who consistently under-price risk