Probabilistic models of financial behavior are growing necessarily more complex given newly available data.
The standard elements of the actuarial toolkit haven't changed so much, rather the technical scope of actuarial analysis has dramatically increased. Classically trained actuaries may not have the appropriate programming and data engineering knowledge to do their best possible work. This is where data science can help.
When modeling investment risk, particularly for lending and insurance, predictive modeling techniques like machine learning must be carefully applied in order to maintain a balance between accuracy and explainability. It may be tempting to train the most accurate possible model using an ensemble of black box machine learning methods, but underwriters often require a certain level of explainability to ensure opportunities are being weighed fairly against one another.
In this golden age of data visualization, we are constantly pushing the limits of what can be considered explainable. Complicated models like gradient boosting trees or ensembles between categorical and numerical models can be effectively visualized using tools the D3. Greater modeling complexity gives financiers an edge against competition by allowing them to exploit undercovered sections of their markets.
We frequently work with clients that are aware of the benefits of running a data-driven financing business, but are not quite ready to undertake predictive modeling. Predictive modeling, or machine learning, requires a considerable investment performance history. With these clients, we will do the following.
Scoring systems are as old as time, but they are stepping stones towards machine-learning-driven investing. They are necessary in order to maintain profibility while the investment philosophy and investment data mature. In many lending and insurance practives, if underwriters manually review every deal, they will get swamped and fail to underwrite at sustainable volumes. Scoring systems help increase and maintain investment velocity in such enterprises.
Most insurance companies have a very traditional data collection formula whereby they ask for certain information from the customer through a collection of forms, plug it into an existing risk model, and price the service accordingly. This process works fine in theory, but there are ample opportunities for the customer to mistakenly record or omit important information that may have a hidden effect on the portfolio.
For example, if an insurance company was insuring a residential property, would it not be to the company's benefit to know if the property was on the rental market? Even if the insurance company asks the customer about this matter, should the insurance company still try to scan sites like AirBnB and VRBO just to double check? Further, if the resident was presently the victim of a rental scam, would it not be in the best interest of both the insurer and the insured to know about it?
The above is just one potential example of how investments (insurance or otherwise) can be secured, protected, and monitored through exotic or alternative data collection methods.