Discriminatory integrated filings may be a way of your LTCI mess, but only if you are part of an integrated filing and only if you are hardy.
Recently, I have been asked about the fairness of integrated filings. Transamerica, an LTCI carrier with 50 books, has used an integrated filing (IF) approach since 2009.
What is an integrated filing (IF)?
Nearly all LTCI carriers have more than 1 Book of business. An IF combines several books into one logical filing. By that, we mean the Master Exhibit values represent the sum of premiums & claims across all books. The lifetime loss ratio that drives rate adjudication is calculated accordingly. To be sure, other filings combine books under one cover (SERFF-ID) but the books are logically separate.
Transamerica and Brighthouse are known to use IF. Other carriers may use IF in the future.
Administrative efficiency and lower cost is one reason to use IF. The question I was asked is whether it is discriminatory. Most people would intuitively answer this question “Yes”! Is there a data science technique to answer this question?
First, if every filing for a rate increase has been integrated for over a decade as in Transamerica’s case, the task would be to somehow disassemble the filing into its 50 component books. This is a monumental, mind boggling task. The carriers hold the cards by not releasing details.
Brighthouse in the past two years evolved to an IF from 13 book filings. Key conclusions from a research experiment looking at this case reveal:
- Books that have a low ratio of current rate : SDN (fair rate) just prior to integration are treated very favorably, having escaped a high potential premium had they stayed on their own. These tend to be underperforming books that we sometimes refer to a trash.
- By contrast, books that have a high ratio just prior to integration are treated very unfavorably. Why is this? They are subsidizing the trash to the degree of a trash’s weight.
An integrated filing is a zero sum game for consumers. Unless all (13 or 50) books perform the same, you have winners & losers. If you are a winner, do not rock the boat. If you are a big loser, consider rocking the boat.
Integrated filings are discriminatory
How you do know if you are a loser or winner? Without any data we have a rule-of-thumb: If you had reason to believe you were winning before, you probably just lost. Vice versa. Rule-of-thumbs lack certainty. If certainty is necessary, the required data together with a computational task is required to derive SDN to be conclusive on discrimination and financial loss. We do that for stakeholders when there is an incentive.
Here is where it can get nasty
What about the questioner who wants clarity. Have they been on the losing end subsidizing the trash? What is the loss in hard dollar terms? I say to them, you need to acquire the most recent filing of all 50 books. One could do the same experiment as was done for Brighthouse.
Just ask for the 50 filings. Say you have a right to know. You know that I am kidding, right!?
Another approach is to ask the carrier (or DOI) to disassemble your policy history to determine you have been treated fairly. What is their criteria that your book has not been unfairly treated? Shouldn’t this be stated in the filing already?
Simplest approach: If the carrier (or DOI) is unable or unwilling to track it through, then you may have a claim that your Original Policy Assumption (Methodological Quirk) stands. What would that mean exactly? OPA’s were designed to meet minimum statutory loss ratio. You are back to square 1.0 x your original premium. Forever.
Wouldn’t mind hearing an intelligent rebuttal or two as this is untested.
Final thought: Brighthouse is not the subject of this post. It is the perfect case to respond to a policyholder’s inquiry.
An LTCI filing that includes original assumptions in the Master Exhibit reveals yet another methodological problem in rate adjudication with long-term consequences to policyholders.
This post describes another example of:
We use a 1 page Master Exhibit 2017 SERFF-ID MULF-130960272, a 666 page filing covering several books.
The exhibit contains three 3-column sets:
- Original Pricing Assumptions (OPA) in the 1st 3-column set;
- The experience / projection before EP(b)(b) the requested rate increase as the 2nd column set; and
- The experience / projection after EP(a) the requested rate increase as the 3rd column set.
Each 3-column set includes incurred claims, premiums, and loss ratio (claims / premiums) for each year of the 52 year life (1998-2040). The summary includes the lifetime loss ratio (LLR).
OPA is the initial justification for pricing a policy in the beginning of a book’s life. Consider OPA random fiction.
Note the yearly loss ratio (LR) comparisons of OPA vs. either of the two last 3-column sets (EPs). OPA’s LR is always greater except for 2007 – 2008. Despite this, OPA’s LLR (0.67) is lower than EF(b) (0.92) and EF(a) (0.87). This seems quirky. It would be natural to ask of this riddle…
How Could This Be?
Observe EP premiums & claims are well above OPA starting around 2000. Could the carrier have added books along the way, but continue to use the same OPA? It appears so!
Answer to the riddle: Since EP premiums & claims >> OPA in the weighty years when annual loss ratios exceeds 90% (e.g. years 2005 & beyond), these later years dominate the LLR calculation.
Shouldn’t the industry (carrier, regulators) have known by 2007 that OPA was far off given the wide premium & claim variance between OPA and EP?
What difference does this make?
Refer to the table below. This is the framework used by some regulators to judge how much of an increase is warranted, if any. A comparison is made between actual v. expected loss ratios, the latter from the dated and obsolete OPA!
|Year||Earned Premium (billions)||Incurred Claims (billions)||(1) Actual (LLR)||(2) Expect (LLR)||(1)/(2)|
Using the invalid expected LR(2) for the stated periods (1991-2013, 2014, 2015) meant the carrier’s rate requests had to meet a secondary test in addition to the primary minimum statutory (MLLF of 0.60), thus causing a delay in rate increases. Such delays exascerbate the well-known Catch Up problem where future increases must make up for lost time. Policyholders presently own this liability though we make the case that they should not.
My guess is that the LTCI industry is not even aware of this problem or its consequences.
Other cases (e.g. AEGB-131679838) use an OPA that has the opposite problem. Here, we see OPA LR(s) consistently under EP LRs and OPA. This means the secondary test would not block rate request increases that would be otherwise blocked.
Rate grants to your policy are sensitive to whatever fictional OPA created over two decades ago.
Final Note (MULF-130960272): We used our app to review OPA and EP key metrics. The Step Down (SDN) for the OPA is 1.16>1.0 since the OPA had an LLR of 0.67 as opposed to the 0.6 MLLR. By contrast, the EP filings have an SDN of 1.73 reflecting a rise in the claims-to-premium ratio compared to OP. The Step Up (SUP) are 12.5 and 14.5, respectively for OPA v. EP. In either case, both books are Catch Up sensitive. Rate increases will continue as they have since the first in 2009, now a total of five. The EP increases were not granted in a timely fashion, due both to a dozing industry and secondary test described above. The current premium is 2.36x the base aggregate premium (1.0), +26% above the EP SDN (1.73). Policyholders are currently paying a 26% subsidy of their current annual premium to pay for past losses.
LTCI has suffered for lack of risk-based long-term horizon scanning to the detriment of its clients. This is first in a forensic series of posts.
In the best case, the LTCI regulatory apparatus has proven itself inadequate concerning principles of risk-based long-term horizon scanning. A worse case would be if the carriers’ scheme was intentional and sufficiently subtle to dupe the regulatory agencies. The worst case is that neither carrier or regulatory apparatus is aware of long-term horizon scanning.
Risk-based horizon scanning
It is a formal effort and ability to identify risks or assumptions that could go wrong to the best plan. Often, you hear the industry’s standard lament (excuse) for rate increases: (1) interest rates too low, (2) claims projections were understated, and (3) inadequate lapse rates. May I ask whether there were any sensitivity analyses performed as is customary for any mathematical modeling endeavor? Stress tests? Apparently not. This goes in the book as a major product design defect for failing to meet standards of basic commercial modeling.
Our recent efforts in Time Series Analysis (TSA), a forensic tool, have been revealing. Our first post on TSA was LTCI Time Bandit which explains some key metrics (SUP, SDN, CEP) used below. The case below makes a point about long-term horizon scanning.
The experiment below concerns the rate filing history of carrier AEG though it can be any carrier (book). Since 2012, numerous AEG filings have lifetime loss ratios lingering in the range of 102% – 118%, with 106% being the most recent. Sustaining this range when the minimum statutory lifetime loss ratio (MLLR) is 60% (or 80%) can guarantee year after year filings for rate increases as AEG has done — 10 rounds of increases since 2009 many on the mild side 10% or 15%, sometimes overlapping. Cumulatively they amount to ~3.3x original premium. What a business!
For this experiment, we took a 2012 rate filing (AEGJ-128207180), cloned it, and gave it an artificial filing date of 2020. Both were given the same rate history. Applying our Consumer View app, we find both the original and the clone have a Step Down (SDN) of ~1.92, a validation of the cloning procedure. SDN assumes: PHs are not responsible for past losses; identifies a level premium given carrier’s claims history and expense projections (CEP); the so-called fair premium is a multiple of a PH’s original premium.
The 2020 clone Step Up (SUP) was 11.96 : 5.09 or 2.35x the 2012 filing (SUP). The corresponding annualized compounded increase necessary in 2020 to achieve the minimum statutory lifetime loss ratio (MLLR) is 20% whereas only 14% in the 2012 case.
Policyholder’s vulnerability to rate increases in the 2020 clone case has nothing to do with the industry’s standard lament! It is exclusively due to a dwindled PH count, shouldered with the burden to correct the book to the MLLR in a shortened time window. If we were to clone 2012 as a 2025 case, the results would be significantly worse.
The current flawed LTCI modeling framework is a great scheme for carriers, even if they do not perform risk-base horizon scanning. Why make the effort? More work and they are already winning. Not such a great scheme for regulators/NAIC, whether or not they are aware of it.