ISO Emerging Issues Perspective
By Jared Smolik
REFINING AUTO RATING WITH DATA ANALYTICS
New tools help insurers align premium with risk
Risk factors old and new are coming into sharper focus as personal auto insurers adapt to emerging data streams and technologies. The implications could filter all the way down to the retail level, with new research leading to greater understanding and more refined pricing in two areas of growing interest: mileage and advanced driver assistance systems (ADAS).
There’s nothing radical about mileage as a rating factor; it’s long been recognized as highly predictive for risk of loss. The effects of ADAS, however, are just beginning to emerge as the technology develops and infiltrates the automotive fleet. What mileage and ADAS have in common is elusiveness, albeit for different reasons.
For as long as insurers have recognized the value of knowing mileage, capturing that data has largely been subject to self-reporting and guesswork—only some of it educated.
For as long as insurers have recognized the value of knowing mileage, capturing that data has largely been subject to self-reporting and guesswork—only some of it educated. Agents often find themselves between the insured and the insurer in a cat and mouse game that may or may not lead to accurate numbers.
ADAS, meanwhile, is a multilayered problem. There are questions of identifying which vehicles have what systems, how different systems affect accident frequency and severity, and whether the benefits outweigh the expense of repairing pricey ADAS hardware after an accident.
Fork in the road
Differences between reported and actual mileage are likely as old as auto insurers’ efforts to collect this data. Agents and insurers have long relied on applicants and policyholders to report their annual mileage, but many customers under-estimate or understate the number. They may simply not know, or they may be seeking to lower their rates. Either way, the result is unreliable data on a rating factor that, if captured accurately, has immense value to insurers.
This conundrum has led some insurers to assume a minimum annual level, such as 3,000 miles (about 10 miles per weekday), when determining premiums. Others employ wide mileage rating bands or exclude mileage from their rate plans altogether. But new research from Verisk helps illuminate the scope of the mileage issue.
Verisk’s 2018 Verified Mileage Study found that about two-thirds of drivers underreported annual mileage to their insurers, and more than 25% of estimates in the study were at least 6,000 miles less than the actual figures. That means an insurer using 3,000-mile rating bands may be mispricing by two or more bands on a sizable portion of its portfolio. Average annual underreporting was 3,177 miles. The study looked at about 153,000 vehicle identification numbers (VINs) for which Verisk had access to both verified odometer readings and vehicle owners’ self-reported mileage estimates, the latter obtained through ISO Statistical Plan filings.
It’s worth noting that in a 2012 survey by the Coalition Against Insurance Fraud, 16% of consumers said they considered it acceptable to lie to an insurer about miles driven. Compounding the issue, one-time events can cause isolated mileage spikes, while changes in residence, employment, or covered vehicles also can nullify historical odometer readings. The implications aren’t academic. Verisk estimates that underreported mileage drives more than $5 billion—or about 18%—of annual premium leakage.
But there’s some good news. Previous Verisk research showed how more and narrower mileage bands could refine underwriting: Vehicles clocking no more than 3,000 miles annually had 44% fewer
claims than the average for all vehicles studied. Conversely, at more than 20,000 miles a year, vehicles had 28% more claims than the average. Armed with this knowledge, insurers can attack the mileage problem knowing that it’s likely to be a worthwhile effort. In addition to refining the bands, here are some ways forward:
- Telematics data—collected with consumer consent through connected vehicles, plug-in devices, or mobile apps—can support pay-per-mile usage-based insurance (UBI) programs. This approach can put mileage front and center in auto programs while eliminating the struggle and uncertainty over accuracy.
- State inspections, auto maintenance service providers, and dealership service departments can capture odometer readings to provide verified mileage at regular intervals.
- Where odometer data is scant or nonexistent, such as for recently purchased cars or older vehicles with less advanced technology, credible estimates can be modeled using geographical and household data.
New ISO rating factors for mileage based on verified odometer readings are currently slated for filing with state regulators in 2019.
How much safer?
While some new tools can improve how mileage data is acquired, other technological advances affect how safely the miles accumulate. ADAS is clearly affecting vehicle risk, but data is just starting to emerge that can help insurers improve rating segmentation and better align premium with risk. Insurers need to know how ADAS affects costs because these systems improve the chances of avoiding crashes but potentially drive up the price of repairs when accidents do occur.
ISO’s personal lines actuaries analyzed extensive premium and loss data from the voluntary market, including vehicle build sheet data, to measure how ADAS affects vehicle risk. Four groupings were created to encompass individual ADAS features, based on how the technology interacts with the vehicle and the driver:
- Passive technologies operate to help maintain control or stability without the driver’s feedback. Driver actions—such as hard braking that activates maximum braking force via the vehicle’s brake assist feature—can trigger these systems. A vehicle’s internal sensors, such as a pitch sensor that detects the tilt of the car’s body, also can activate these features.
- Driving feedback contains technologies that audibly and/or visibly alert the driver. Sensors and/or cameras monitor the vehicle’s surroundings and activate warnings that can prompt driver reaction to avoid or mitigate a collision.
- Parking feedback provides audible and/or visible guidance to the driver when parking. Like driving feedback, these features monitor the vehicle’s surroundings with sensors and/or cameras and warn the driver of potential collisions.
- Active technologies monitor the vehicle’s surroundings with sensors and/or cameras and will apply turning or braking inputs independent of the driver in an attempt to avoid a collision.
As vehicle autonomy advances, rating of vehicles will present insurers with new challenges. For example, software may drive many autonomous features; therefore, the software version that a vehicle is running may be as important as what hardware is onboard.
ISO staff’s findings suggest that particular ADAS groupings result in discounted rating factors being applied to certain coverages. Discounts for third-party coverages generally are higher than for collision coverage, and the greatest impact is on property damage coverage. Industry research has found that severity of collision losses can offset, in whole or in part, the drop in frequency because ADAS technologies drive repair costs higher.
ISO staff projects collision experience to improve as ADAS features penetrate more of the vehicle fleet, while frequency of accidents is forecasted to continue declining. ISO began to file rating factors for the ADAS rating groups described above in the first quarter of 2019.
Development of new tools for mileage capture and uncovering the implications of ADAS features are two examples of the growing power of data analytics—to shine fresh light on historical problems and give the industry a better handle on new challenges. The ability to apply rating factors based on newly minted data shows that solving these issues may be a stretch for insurers, but answers aren’t out of reach.
Jared Smollik is vice president of personal lines actuarial and analytic products at ISO, a Verisk (Nasdaq:VRSK) business.