Dislocation Analysis:
Segment Detection of Disrupted Rates
March 2022
Rob Zolla, FCAS, MAAA Guillaume Béraud-Sudreau
Dislocation Analysis:
Segment Detection of Disrupted Rates
Rob Zolla, FCAS, MAAA
March 2022
Rob Zolla, FCAS, MAAA
Rob is an Actuary at Milliman’s Orange County office
specializing in InsurTech solutions and advanced
ratemaking techniques for personal and commercial lines.
He has over 10 years of experience in the insurance
industry, much of which came at a top five U.S. personal
lines carrier where he held executive roles leading teams
focused on competitive intelligence, property pricing, and
product and revenue forecasting.
3
Poll 1
What is your current position?
A. I am an individual contributor with respect to the rate change
process.
B. I am in a leadership position making decisions regarding rate
changes.
C. I work for a regulatory body reviewing rate change submissions.
D. None of the above.
4
Poll 2
What line of business do you primarily focus on?
A. Personal Lines Ratemaking
B. Commercial Lines Ratemaking
C. Combination of Personal/Commercial Lines Ratemaking
D. Non-Ratemaking
5
Poll 3
How would you rate the analytical capabilities available to you to perform or review dislocation
analyses?
A. 1 – Best in Class
B. 2 – Above Average
C. 3 – Below Average
D. 4 – Poor
E. Not Applicable
6
Initial Considerations
Adhering to Actuarial Standards of Practice, Laws, and Regulations
▪ ASOP 53 – Estimating Future Costs for Prospective Property/Casualty Risk Transfer and Risk Retention
▪ Section 3.1 Future Cost Estimate – “The actuary should determine the elements that are appropriate to include in the
future cost estimate. Such elements should relate to the applicable coverage and include loss and loss adjustment
expenses, operational and administrative expenses, the cost of reinsurance, and the cost of capital.”
▪ ASOP 56 – Modeling
▪ Section 3.1.2 Selecting, Reviewing, or Evaluating the Model – “When selecting, reviewing, or evaluating the model,
the actuary should confirm that, in the actuary’s professional judgment, the model reasonably meets the intended
purpose.”
▪ Section 3.6.d Evaluation of Mitigation of Model Risk – “The type and degree of model risk mitigation that is reasonable
and appropriate may depend on the following…..whether there have been changes to the model or its operating
environment.”
▪ ASOP 30 – Treatment of Profit and Contingency Provisions and the Cost of Capital in Property/Casualty Insurance
Ratemaking
▪ Section 4.1 Conflict with Law or Regulation – “If a law or regulation conflicts with the provisions of this standard, the
actuary should develop a rate in accordance with the law or regulation, and disclose any material differences between
the rate so developed and the actuarially determined rate to the client or employer.”
7
Dislocation Impacts
Dislocation Impacts on CAS Exams
▪ “Typically, companies look at the
distribution of rate changes across the
entire book of business, summarized by
key segments, and by each level of rating
variables being specifically adjusted.”
▪ “Once the traditional actuarial indications
and marketing considerations are known,
the decision-maker needs to weigh all
information and select the rates that best
meet the goals of the company. Typically,
this is done judgmentally.”
Source: Werner, G, and Modlin, C., Basic Ratemaking, Casualty Actuarial Society, Fifth Edition, May 2016.
8
Casualty Actuarial Society
2021-23 Strategic Plan
▪ The inspiration for the Strategic Plan is illustrated through a new Envisioned Future that foresees:
▪ CAS members are sought after globally for their insights and ability to apply analytics to solve insurance
and risk management problems.
▪ The Strategic Plan is designed to prepare CAS members with skills in three primary areas:
1. Analytics, to tackle the important insurance and risk management problems in our data-rich world.
2. Problem solving, built upon strong strategic thinking and communication skills.
3. Domain knowledge, specifically property and casualty insurance and risk management.
Source: CAS Strategic Plan | Casualty Actuarial Society (casact.org)
Casualty Actuarial Society
2021-23 Strategic Plan
▪ “It is becoming increasingly important to have technical and soft skills to translate analysis into business
terms. While technical skills are needed to get a foot in the door, those best positioned for success will also
possess.”
▪ The intellectual curiosity to proactively identify business problems.
▪ The ability to apply logic to a problem.
▪ The business and industry knowledge need to provide effective recommendations from the analysis.
▪ Strong communication skills to tell a clear story.
▪ The ability to conquer the “last mile” problem in analytics to successfully get end-users to use the
analytics for impact.
10
The Pricing Actuary and Function Integration
Applying Analytics to Solve Insurance and Risk Management Problems
• Prioritization of analyses • Impacts to risk
• Regulatory/data/financial metrics
constraints • Regulatory approval
• Analytical constraints
Data
Risk/
Science/
Regulatory
Modeling
Product
Mgmt/ Finance
Agents
• Growth goals • Translate ratemaking
• Policyholder impacts impacts to financial
• Impacts from prior rate impacts
changes
11
Pre-Work for an Analysis
What Does the Analysis Set Out to Achieve?
• Profitability, Growth, Subsidy
Current State • Marketplace Trends
• Recent Rate Changes
Residual Impacts • Product/Agent Feedback
• Time available
Resources • Budget considerations
• Shareholders or policyholders
Stakeholders • Internal stakeholders
12
Dislocation Considerations
How Companies May Approach Different Scenarios?
Scenario 1 Scenario 2
▪ Mature program ▪ Newer program
▪ Program with sustained profitability ▪ Program with profitability challenges
▪ Sustainable growth ▪ Lack of growth in key segments
Additional Considerations:
▪ Where is the insurance market at in the underwriting cycle?
▪ What are the pricing actions of competitors?
▪ Are there market segments with a history of rate increases?
▪ Should policies in force be expanded to better reflect the target market?
▪ What are potential non-pricing actions?
13
Regulatory Constraints
Understanding Regulatory Requirements Prior to Submitting a Filing
▪ General Restrictions:
▪ California – Proposition 103
▪ “The application shall be deemed approved sixty days after public notice unless the proposed rate adjustment exceeds 7% of the
then applicable rate for personal lines or 15% for commercial lines, in which case the commissioner must hold a hearing…”
▪ New York – 30% maximum policy-level rate change
▪ Alabama – Filing frequency restricted to coincide with the typical term of coverage
▪ Restrictions on Rating Variables:
▪ Washington – Credit Scoring
▪ “Commissioner Kreidler adopted his rule temporarily banning credit scoring.”
▪ California – Gender (Personal Auto)
▪ Tennessee – Fictitious Grouping
▪ “Preferences or distinctions in certain insurance transactions prohibited.”
▪ “Fictitious grouping” means any grouping by way of membership, nonmembership, license, franchise, employment, contract,
agreement or any other method or means;
14
Implementation Considerations
▪ Methods for Limiting Disruption:
▪ Individual Rating Factors – Review selections
between current factors and indicated factors.
▪ Rate Capping/Rate Stability – Review state
regulatory requirements.
▪ Advancement Rate Capping/Rate Stability –
Asymmetrical capping or capping by renewal
period.
▪ Non-Pricing Actions:
▪ Underwriting
▪ Marketing
▪ Fraud detection
▪ Premium leakage
15
Communicating to Internal Stakeholders
Results
Internal Support
• Product • Timing of Monitoring
• Marketing • Gather broad benefits • Timing and
• Claims internal • Interpretability approach for
• Analytics champion of results monitoring
support
Financial and
Gather Feedback
Customer Impacts
16
Dislocation and Disparate Impacts
Additional Considerations
▪ “Minority Neighborhoods Pay Higher Car Insurance Premiums
than White Areas With the Same Risk” – ProPublica 2017
▪ “Confronting the Issues of Race and Pricing” – Actuarial Review
▪ Possible ways to examine price discrimination.
▪ Do nothing.
▪ Exclude certain risk classifications from rating plans.
▪ Introduce a variable that controls for a protected risk characteristic.
▪ Evaluate the final price’s impact on protected classes.
▪ “We’re dropping credit score from car insurance pricing by 2025”
– Root Insurance Blog, 2021
17
Thank you.
Rob Zolla, FCAS, MAAA
March 2022
Dislocation Analysis:
Segment Detection of Disrupted Rates
2022
Biography
Guillaume is the Chief Actuary and
Co-Founder of Akur8.
He has both a data science and an
actuarial background.
Guillaume started researching the
potential of AI for insurance pricing as
Head of Pricing R&D at Axa Global Direct,
before being incubated at Kamet
Ventures and founding Akur8.
Guillaume Béraud-Sudreau Guillaume is a Fellow of the
Chief Actuary & Co-Founder of Akur8 French Institute of Actuaries and holds
Master’s degrees in Actuarial Science,
Cognitive Science and Engineering from
Institut des Actuaires - CNAM, Ecole
normale supérieure, and Télécom Paris.
Pricing update process
Indicated Rates Proposed Dislocation
Premiums Adjustments Premiums Analysis
Validation &
Future best-estimate of the
policies risks
Future premium candidates Production
in-production
Final validation of the new pricing
Adjust rates to integrate Current structure and deployment in production.
product considerations
Premiums
Previous premiums currently Pricing comparisons and
used in-production analysis of the rate changes
CONFIDENTIAL 21
Premiums Comparison
The Basic: Price-Change distribution
The simplest approach only provides a view of the price changes.
This graph also provides a view of the importance of the proposed price change.
CONFIDENTIAL 23
The Basic: Price-Change distribution
Significant Decrease: Stable Rates Significant increase:
loss of opportunity High risk of lapse (and potential
regulatory limits)
CONFIDENTIAL 24
For instance: 3 strategies (1/3)
No Constraints
There is no specific constraints in terms of dislocation; e.g., the current portfolio is either very small or
extremely unprofitable, and the new indicated premiums can be considered as the only relevant information.
CONFIDENTIAL 25
For instance: 3 strategies (2/3)
Soft Update
Keep dislocation small to limit the disruption; e.g., for large and/or profitable portfolios, the new indicated
premiums aim to control arising loss patterns, however keeping rates stable is the priority. As a result, a desired
strategy is to adopt a soft update with small price changes.
CONFIDENTIAL 26
For instance: 3 strategies (3/3)
Capped Updates
For either regulatory or operational reasons, the updates are limited to a certain range (upper limit of 30% in
the example) to avoid extreme dislocations.
CONFIDENTIAL 27
Who has been updated?
Density of Current vs Proposed price
Premium Density
Dark-red areas indicate a high density of
policies; lighter yellow areas represent
lower density.
CONFIDENTIAL 28
Who has been updated?
Density of Current vs Proposed price
Premium Density
Price Increase ⬆ Same price
Price increases and decreases are
clearly identified on the graph.
Policies along the diagonal are not
impacted by the rate change.
Price Decrease ⬇
CONFIDENTIAL
29
Premium Density
+50%
Who has been updated? Same Price
Density of Current vs Proposed price
The Premium Density graph provides a -35%
richer view of the price changes
distribution.
E.g., people with a given price increase
(one point of the Rate Change graph)
will be on the same line in the Premium
Density graph.
-50% 0% +50%
CONFIDENTIAL 30
Rate Change
Small or large price update?
1 No Constraints 2 Soft Updates 3 Capped Updates
In the Capped Updates scenario (right), all the points are grouped around the diagonal where the new proposed price equals
the current one. On the No Constraints scenario (left), the price changes are wider: some points are far away from the
diagonal indicating the new proposed prices are very different from the current prices.
CONFIDENTIAL 31
Who was impacted, and how?
Variable-by-variable review of the price change
The Vehicle_Value_at_Purchase variable has the most significant impact on the price change: any review of the
new price should involve an in-depth study of the relationship between this variable and the observed risk.
CONFIDENTIAL 32
Who was impacted, and how?
Variable-by-variable review of the price change
Relative impact of Segmentation of
each variable on the Rate Change
the Rate Change
CONFIDENTIAL 33
Who was impacted, and how?
Variable-by-variable review of the price change
The Vehicle_Value_at_Purchase variable is discriminating less the new proposed premium; customers were
likely overly discriminated in the current premium structure.
CONFIDENTIAL 34
Who is right?
Price Update Analysis
The proposed vs current price
These graphs provide a full view of which profiles are impacted by the price change.
They don’t give information on how well justified these changes are.
CONFIDENTIAL 36
Models Performances
Simply compare the underlying models performances?
Models underlying the Current Premium Model underlying the Proposed Premium
Compare the commercial premiums alignment with the risks.
This approach is purely “performance-based”, according to the data available.
CONFIDENTIAL 37
Models Performances
Compare the “premium performances”?
Compare the commercial premiums alignment with the risks.
This approach is purely “performance-based”, according to the data available.
CONFIDENTIAL 38
Loss-Ratio and Premiums
Are there premium segments with “wrong” loss ratios?
A loss ratio comparison by premium
segments allows insurers to spot trends
and anomalies, for example: the proposed
premium under-prices the low risks (first
deciles).
CONFIDENTIAL 39
Mixing the change and performance analysis
Compare performances?
Enriching the 2D premium distribution graph with the loss ratio provides a full vision of the premium
changes and the “quality” of the policies impacted.
CONFIDENTIAL 40
Mixing the change and performance analysis
Compare performances?
High Price Increases
& High Loss Ratios
Is the price increase motivated?
CONFIDENTIAL 41
Mixing the change and performance analysis
Loss Ratios, Price Changes and Constraints?
Price increases
for high LR?
Price decreases
for low LR?
If contracts with price increases also have high LRs, maybe we should increase the prices even more?
CONFIDENTIAL 42
Scenario 1: No constraints
Smoothed Loss Ratio
The update aims for an homogeneous
loss-ratio.
The LR distribution is expected to be narrow,
but at the expense of the premium
consistency.
CONFIDENTIAL 43
Scenario 2: Soft Update
Smoothed Loss Ratio
The price-update is limited:
● Identified high risks are “under-increased”:
they are red on the graph
● Identified low risks are “under-decreased”:
they are blue on the graph
This graph has a rainbow appearance.
CONFIDENTIAL 44
Scenario 3: Capped Updates
Smoothed Loss Ratio
The points within the +/-30% range are not
impacted by imposing the limits: these loss
ratios are the same as the loss ratios in the No
Constraints scenario.
CONFIDENTIAL 45
Performance: Are changes motivated?
High loss ratios are expected to be on contracts whose
prices have been increased (partially increased)
All loss ratios that do not follow the rainbow pattern
(premiums of high risks are increased, premiums of
low risks are decreased) are caused by noise, errors, or
non-technical price adjustments (for example involving
external constraints or other product decisions).
CONFIDENTIAL 46
Performance: Are changes motivated?
We want to quickly identify who is in the investigated
segment:
● Review which variables discriminate between
contracts that are / aren’t within the segment
● For each impacted variable, review which levels
are in/out of the segment (in the example, the
identified segment mainly includes light
vehicles)
High LR & price decrease??
CONFIDENTIAL 47
Who is this?
We want to quickly identify who is in the investigated segment:
● Review which variables discriminate between contracts that are / aren’t within the segment
● For each impacted variable, review which levels are in/out of the segment (in the example, the
identified segment mainly includes light vehicles)
CONFIDENTIAL 48
Who is this?
We want to quickly identify who is in the investigated segment:
● Review which variables discriminate between contracts that are / aren’t within the segment
● For each impacted variable, review which levels are in/out of the segment (in the example, the
identified segment mainly includes light vehicles)
CONFIDENTIAL 49
CONFIDENTIAL 50
Questions?
Thanks!
Guillaume Béraud-Sudreau
Chief Actuary & Co-Founder of Akur8
[email protected]
