Materials - Investment Risk-Based Capital (E) Working Group

Date: 11/21/16 2016 Fall National Meeting Miami, Florida

INVESTMENT RISK-BASED CAPITAL (E) WORKING GROUP Sunday, December 11, 2016 8:00 – 9:00 a.m. Fontainebleau Miami—Glimmer 3-4—Level 4

ROLL CALL Kevin Fry, Chair Philip Barlow, Vice Chair Greg Lieber Kerry Krantz Kathy Belfi

Illinois District of Columbia California Florida Connecticut

Chris Buchanan Anna Taam Steven Drutz Richard Hinkel

Kansas New York Washington Wisconsin

AGENDA 1.

Consider Adoption of its Minutes—Kevin Fry (IL) • Sept 8 • Oct. 20

Attachment A Attachment B

2.

Hear Presentation from the American Academy of Actuaries (Academy) Related to Portfolio Adjustment for Bonds—Nancy Bennett (Academy) and Rich Owens (Academy)

Attachment C

3.

Continue Discussion on the Bond Factors in the Life Risk-Based Capital (RBC) Formula —Kevin Fry (IL)

Attachment D

4.

Continue Discussion on Whether the Bond RBC Structure Should Be Consistent Across All Statement Types—Kevin Fry (IL)

5.

Discuss Any Other Matters Brought Before the Working Group—Kevin Fry (IL)

6.

Adjournment—Kevin Fry (IL)

W:\National Meetings\2016\Fall\TF\CapAdequacy\InvestmentRBC\National Meeting\IRBC WG Agenda Fall 2016.doc

© 2016 National Association of Insurance Commissioners

1

1

2

Attachment A Attachment XX Capital Adequacy (E) Task Force 12/xx/16 Draft: 9/19/16 Investment Risk-Based Capital (E) Working Group Conference Call September 8, 2016 The Investment Risk-Based Capital (E) Working Group of the Capital Adequacy (E) Task Force met via conference call Sept. 8, 2016. The following Working Group members participated: Kevin Fry, Chair (IL); Philip Barlow, Vice Chair (DC); Greg Lieber (CA); Kathy Belfi (CT); Shawn Steinly (FL); Chris Buchanan (KS); Anna Taam (NY); Steven Drutz (WA); and Richard Hinkel (WI). Also participating was: Dale Bruggeman (OH). 1.

Discussed Involvement of Other Working Groups Under Capital Adequacy (E) Task Force

Mr. Fry said the Working Group was originally named the C-1 Factor Review (E) Subgroup and was charged with reviewing investment-related issues in only the life risk-based capital (RBC) formula. Over time, the Working Group’s charges evolved, and it is now responsible for considering investment-related issues for all of the RBC formulas. This is a unique situation, because there are also other working groups under Capital Adequacy (E) Task Force whose primary charge is to maintain these formulas. Because of this, there is a natural amount of overlap will take place. Questions have been raised related to how and when the other working groups—life, health and property/casualty (P/C)—will be involved in the investment RBC project and what their role will be. Mr. Fry said the Working Group needs to work on its charge in an efficient and expedient manner, while also ensuring that the appropriate subject matter experts and interested stakeholders are sufficiently involved. To that end, whenever the Working Group exposes a document for public comment, the exposure email will be sent to not only the distribution list for this Working Group, but it will also be sent to those on the distribution list for the working group responsible for that particular RBC formula. For example, if a document is exposed that includes revisions the life RBC formula, the exposure email will be sent to the distribution lists for both this Working Group and the Life Risk-Based Capital (E) Working Group. In addition, during a conference call in which comment letters are discussed, the call will be a joint conference call between this Working Group and the other affected working group(s). This approach ensures there is involvement by the other working groups with responsibility for the RBC formulas, but the process will not be complicated with numerous working groups discussing the same topic at different times. 2.

Exposed Revisions Related to the Bond Structure in the Life RBC Formula

Mr. Fry said that, for the past few months, the Working Group has been discussing the life bond structure and has reached a general consensus on the key considerations on this topic. Up to this point, the discussions have been primarily conceptual, but, during the Working Group’s July 29 call, NAIC staff was instructed to draft the revisions to the life RBC blank and instructions for the expansion of the life bond structure based on the Working Group’s discussions to date. Julie Garber (NAIC) discussed the proposed revisions (Attachment X, Attachment X and Attachment X) and provided an overview of the changes being suggested on each page of the RBC blank, as well as suggested referrals to other groups. She said an issue was identified with regard to hybrid securities. Currently, the bond amounts on page LR002 are calculated by using the amounts in the Asset Valuation Reserve (AVR) Default Worksheet and subtracting the related amount of hybrid securities in each category as reported in Schedule D, Part 1A, Section 1. This subtraction is necessary because the AVR Default Worksheet bond amounts currently include hybrid securities, but the RBC formula calculates the RBC charge for hybrid securities on page LR005, as opposed to page LR002. The problem is created because if the number of bond categories is expanded to 20, the amounts for the hybrid securities on Schedule D, Part 1A, Section 1 only includes the six NAIC designations. Even though hybrid securities are included on a separate page in the RBC formula, they are currently assessed the same RBC charge as bonds based on the six NAIC designations. Ms. Garber said the memo includes two different options to address the issue related to hybrid securities, and NAIC staff would appreciate feedback on these two options. Mr. Barlow asked if the American Academy of Actuaries (Academy) reviewed hybrid securities in its development of the August 2015 report on bonds. Nancy Bennett (Academy) said the Academy did not specifically consider hybrid securities in its analysis. The model and related assumptions were based on corporate bonds, while its consideration of other securities that get the bond treatment was based on a relative risk analysis. The Academy did not believe that any further adjustments were necessary in regard to hybrid securities. Ed Toy (NAIC) said that, at one point, hybrid securities were treated like © 2016 National Association of Insurance Commissioners

1

3

Attachment A Attachment XX Capital Adequacy (E) Task Force 12/xx/16 preferred stock but, considering the way in which rating agencies rate hybrid securities, the treatment was later changed so that hybrids receive the same RBC charge as bonds. He said that the market for hybrid securities and the insurance industry’s exposure to them has gotten considerably smaller in recent years. Because of this reduced exposure, it may no longer be necessary or appropriate to include hybrid securities in their own line in the RBC blank. John Bruins (American Council of Life Insurers—ACLI) asked for clarification on what the new electronic-only column in Schedule D will look like. Mr. Fry said the Working Group has discussed potentially adding another column to identify certain securities such as municipal and sovereign bonds. Ms. Garber said the electronic-only column would simply refer to the new RBC factor category, and the instructions would include information related to assignment of the category. She noted that the assignment of the RBC factor category will be discussed further when the Working Group begins its consideration of the factor charges. Regarding municipal and sovereign bonds, she said the Working Group will first need to clearly define these securities and then NAIC staff will consider how best to identify them in Schedule D. The Working Group agreed to expose the proposed revisions for a public comment period of 30 days. Although the Working Group has reached a consensus on the key considerations, Mr. Fry said there are a lot of moving pieces in the RBC formula and he expects that there will be at least a couple of iterations of these changes before the Working Group refers a final product to the Capital Adequacy (E) Task Force. 3.

Discussed the Bond Structure in the Health and P/C RBC Formulas

Mr. Fry discussed the deadlines for making RBC-related changes, noting that the deadline for structural changes is earlier than the deadline for changes that only affect the factors. Because of this, he said the Working Group should now focus its attention on the bond structure in the P/C and health formulas, even though the Working Group still needs to discuss the factors in the life formula. Once the Working Group completes its consideration of the bond structure for all RBC formulas, it can then turn its attention to a discussion of the factors. Mr. Fry said the investment schedules in the annual financial statement are uniform across all statement types. Historically, all RBC formulas have used the six RBC factor categories for bonds. The factors in the life RBC formula were based on Academy reports, and the factors were adjusted, as applicable, for use in the health and P/C formulas. Therefore, other than the factors, there was a certain level of consistency among the RBC formulas in regard to bonds. There is now a general understanding that there needs to be more granularity for the bond charges in the life formula. He discussed that, for the majority of P/C and health companies, the bond charges are not typically material in relation to the total amount of RBC. However, there may be some companies in which this may be material. He said the Working Group needs to determine whether there should be more granularity for bonds in the P/C and health formulas. He said that even if the structure for bonds is not changed, the Working Group will need to update the factors based on modern default experience. He said that some have questioned the appropriateness of having differing treatment or charges in the RBC formulas for the same asset. Mr. Bruggeman said that how a life company utilizes fixed income securities is different from how they are used by a P/C or a health company, particularly related to how life companies match cash flows and hold longer-term duration bonds. He noted that the accounting for bonds is different for life companies than it is for P/C and health companies, and an AVR is calculated for life companies. The Working Group should be cognizant of these differences during its deliberation. Mr. Fry said that when the original factors were created, they were adjusted for certain differences such as the accounting treatment. Similar adjustments will need to be made when the Working Group discusses the factors on a future conference call. Mr. Toy said the software vendors have previously been contacted regarding increasing bond granularity, and the feedback he received was that increasing the number of bond categories was not a significant issue for them. He did say, however, that the vendors requested that the same structure be used for all of the formulas, as it would be easier to implement. It has also been discussed that some insurance groups have both life and P/C companies, and different bond RBC structures could create complications for these groups. Mr. Bruins agreed, saying that he would like to see a consistent framework for the RBC bond structure. Mr. Fry noted that, regardless of whether the structure for RBC changes, Schedule D would remain consistent for the different types of filers. Ralph Blanchard (The Travelers Companies) discussed that the same asset could have a different level of risk based on the call risk and whether or when the asset needs to be liquidated. The life formula assumes a certain bond duration that is not necessarily appropriate for non-life companies. Bill Weller (America’s Health Insurance Plans—AHIP) discussed differences that would need to be taken into consideration when comparing life RBC risks for bonds to that of P/C and health companies, including average bond duration, the risk premium included in the statutory reserve for life companies and the accounting © 2016 National Association of Insurance Commissioners

2

4

Attachment A Attachment XX Capital Adequacy (E) Task Force 12/xx/16 treatment. When comparing the life bond model with what would need to be done for P/C and health companies, adjustments would need to be made for these differences. Ms. Bennett said there are two main differences between the current factors for the life formula compared to the health and P/C formulas: 1) taxes; and 2) the carrying value for below-investment-grade securities. In the development of the updated factors, at least two other differences have been discussed: 1) the risk premium included in statutory policy reserves for life insurance; and 2) the time horizon. The risk premium is included in the Academy’s life bond model to recognize the fact that statutory policy reserves for life insurance include provision for expected bond losses. The 10-year time horizon reflects the average credit cycle, and it was not selected to correspond to the average duration of the life insurance liabilities. Ms. Bennett said the Academy's life bond model could easily be adjusted to develop factors for P/C and health insurers by eliminating the tax effects and the risk premium. She noted that while the P/C and health policy reserves do not include a provision for expected losses, the current P/C and health bond factors have not been adjusted for such a provision. If the Working Group wants to explore a different statistical safety level for the P/C and health bond factors with a shorter time horizon (e.g., five years versus the current 10 years), changing the life bond model for this adjustment will be much more time consuming as a shorter time horizon would require the development of different default and recovery assumptions. Below-investment-grade bonds (currently those with a designation of NAIC 3, NAIC 4, NAIC 5 or NAIC 6) are carried at market value in the P/C and health financial statements, while they are carried at amortized cost in the life financial statements. This difference in carrying value needs to be reflected in the factors. Ms. Bennett stated that she does not know how the life factors were originally adjusted for this difference in carrying value and, therefore, is unable to comment on how the life bond model would need to be modified. She said the Academy’s C1 Work Group has already re-run the life bond model with taxes and the risk premium eliminated, and factors have been developed based on these assumptions and the current statistical safety level of 92nd percentile over 10 years. Having no further business, the Investment Risk-Based Capital (E) Working Group adjourned. W:\National Meetings\2016\Fall\TF\CapAdequacy\InvestmentRBC\Sept 8 IRBC CC minutes.docx


3

5

6

Attachment B Attachment XX Capital Adequacy (E) Task Force 12/xx/16 Draft: 11/1/16 Investment Risk-Based Capital (E) Working Group Conference Call October 20, 2016 The Investment Risk-Based Capital (E) Working Group of the Capital Adequacy (E) Task Force met via conference call Oct. 20, 2016. The following Working Group members participated: Kevin Fry, Chair (IL); Philip Barlow, Vice Chair (DC); Greg Lieber (CA); Elaine Wieche (CT); Shawn Steinly (FL); Chris Buchanan (KS); Anna Taam (NY); Steven Drutz (WA); and Richard Hinkel (WI). 1.

Discussed Comment Letters Received Related to the Proposed Revisions to the Life Bond Structure

Mr. Fry said for the last several months, the Working Group has been discussing possible changes to the structure of the life risk-based capital (RBC) formula for bonds, and a general consensus has been reached that the granularity for bonds in the life formula should be increased. On its Sept. 8 conference call, the Working Group exposed proposed revisions to both the life RBC blank and instructions to incorporate the structural changes for bonds in the life formula (see NAIC Proceedings – Fall 2016, Capital Adequacy (E) Task Force, Attachment XX.) Comment letters were received from the American Academy of Actuaries (Academy) (Attachment X) and the American Council of Life Insurers (ACLI) (Attachment X). Nancy Bennett (Academy) summarized the Academy’s comment letter, stating that it is supportive of the Working Group’s proposal to increase the granularity for bonds in the life RBC formula. She said the Academy strongly supports implementing the changes to the base factors and the changes to the portfolio adjustments at the same time. She said that calculation of the C-1 charge is a two-step process in which the first step is application of the base factor, and the second step is adjusting the C-1 charge for the number of bonds in the portfolio. The purpose of this two-step process is to achieve the overall statistical safety level for the C-1 component in its entirety. She said the Academy is in the process of reviewing the current portfolio adjustments and will provide its recommendation to the Working Group in the near future. John Bruins (ACLI) summarized the ACLI’s comment letter, noting it supports expanding the granularity in the life RBC formula and in the calculation of the asset valuation reserve (AVR). He said there are additional details that still need to be addressed such as the format and instructions for the electronic forms, as well as the procedures to categorize assets into the new categories. He discussed continued concern regarding the implementation date of year-end 2017 that has been previously discussed. He also noted that the current proposal does not include information related to health and property/casualty (P/C) companies and does not include separate provisions for certain securities such as municipal and sovereign bonds. Regarding the additional details for the procedures for categorizing assets, Mr. Fry said the proposal discusses a referral to the Valuation of Securities (E) Task Force, so that it may consider how the Working Group’s proposal may operate in light of the National Association of Insurance Commissioners (NAIC) Securities Valuation Office’s (SVO) current operations and procedures. He said the Working Group will continue its discussion on the structure of the health and P/C formulas, as well as the treatment of municipal and sovereign bonds, during a conference call in the near future. Julie Garber (NAIC) discussed some of the specific comments included in the ACLI letter. Regarding specific comment #2, Ms. Garber noted that the subtotals on page LR002 are included for auditability back to Schedule D, Part 1A, Section 1, as well as to provide information for page LR030. From a mechanical standpoint, NAIC staff prefer that these subtotals remain in a row, as opposed to a column as suggested by the ACLI. Regarding specific comment #4, Ms. Garber said that NAIC staff can provide clarifying language in the instructions regarding the bonds to be included on line 22, and agrees with the ACLI that these bonds should be given the top NAIC 1 designation. Regarding specific comment #5, Ms. Garber said NAIC staff agree with the ACLI that Option #1 is preferable and that hybrid securities should be included on page LR002, as opposed to page LR005. Regarding specific comment #9, Ms. Garber conferred with NAIC staff for the Blanks (E) Working Group. Normally, changes that affect both the quarterly and annually statements are typically implemented as of Jan. 1 so that they are first included in first the quarterly statements and then the annual statement. However, the deadline for any change for the 2017 quarterly statements has already passed. Ms. Garber said that although it is not typical, it would not be overly problematic to implement the change to AVR first for the 2017 annual statements and then the 2018 quarterly statements. The Blanks (E) Working Group could provide reporting guidance if there is any confusion on this matter. Mr. Fry suggested that NAIC staff provide a referral to the Valuation of Securities (E) Task Force asking it to review the Working Group’s proposal and consider how the details would work in light of the current operations and processes of the NAIC Securities Valuation Office. At some point, the Working Group also will need to send a referral to the Blanks (E) © 2016 National Association of Insurance Commissioners

1

7

Attachment B Attachment XX Capital Adequacy (E) Task Force 12/xx/16 Working Group regarding proposed changes to Schedule D and the AVR Worksheet, but the Working Group first needs to consider the response from the Task Force before it can make a referral to the Blanks (E) Working Group. Mr. Fry also suggested that a referral be sent to the Statutory Accounting Principles (E) Working Group. Although he said he does not believe the Working Group’s proposed revisions affect the Accounting Practices and Procedures Manual (AP&P Manual), the Statutory Accounting Principles (E) Working Group should confirm this. There were no objections to sending the referrals. 2.

Received a Report from the Academy on its Response to Comment Letters Received on its August 2015 Report Related to the Treatment of Bonds in the Life RBC Formula

At the 2015 Summer National Meeting, the Academy provided an overview of its report “Model Construction and Development of RBC Factors for Fixed Income Securities for the NAIC’s Life Risk Based Capital Formula.” The report recommended increased granularity in the bond base factors for life companies and proposed new factor charges. During the meeting, the Working Group exposed the Academy’s report for a 45-day public comment period, and comment letters were discussed during an Oct. 27, 2015, conference call and the 2015 Fall National Meeting. In response to issues raised in the comment letters, the Academy provided the Working Group with a letter that discusses several general topics and includes answers to certain specific questions (Attachment XX). Ms. Bennett summarized the Academy’s response, noting that the comment letters primarily addressed four basic categories: 1) use of greatest loss versus cumulative loss; 2) definition and quantification of the risk premium offset; 3) discount rate assumption; and 4) applicability of factors to other asset classes. The first two items deal with the methodology of the Academy’s bond modeling process, whereas the last two deal more with the assumptions used in the model. The first category relates to the greatest loss versus the cumulative loss and deals with the basic coverage level for the capital requirement used in the modeling methodology. The issue is whether the RBC requirement should fund the greatest loss at any time during the 10-year time horizon or should it fund any loss at the end of the 10-year time period. In its bond modeling, the Academy used the greatest loss at any time during the 10-year period, and this is consistent with the approach taken when the factors were first developed. The rationale for this is that a company needs to be solvent at all times and not just at the end of a 10-year period. Ms. Bennett said this approach was previously reaffirmed by the Working Group, and she said the Working Group did not feel that any changes to this approach were necessary. The second category relates to the risk premium offset. The C-1 component in the life RBC formula is based on the presumption that statutory policy reserves for future policy benefits are sufficient. Therefore, the C-1 factors establish capital requirements for losses in excess of what is included in the statutory policy reserves. Ms. Bennett said the general actuarial consensus is that life policy reserves make provision for risks under moderately adverse conditions, and RBC establishes funds to cover risks in excess of risks occurring in moderately adverse conditions. The risk premium is the assumption in the bond model that represents the level of losses covered by statutory policy reserves, and the Academy defined the risk premium to be equal to the mean of the loss given default, or roughly the 50th percentile. This is the same approach that was previously used in the modeling methodology, and the Academy found no compelling reason to change it. Some have argued that this definition is overly conservative and that the risk premium should be closer to the 67th or 70th percentile. This approach was previously discussed with the Working Group, and it agreed to use the 50th percentile. The third category is the discount rate, and Ms. Bennett said the C-1 factor is based on the present value of the projected cash flows. The discount rate currently being used is 6% after-tax and was developed in the early 1990s. For the Academy’s 2015 recommendation, it used a 5% pre-tax rate, which is 3.25% after-tax, and this was based on the average 10-year swap rate. Because the C-1 factors are based on discounting after-tax cash flows, she said use of an after-tax rate is appropriate. Further, some questioned use of a risk-free rate versus an earned rate. The Academy used a risk-free rate, and if it were to use an earned rate, certain relatively complex adjustments would need to be made to the model. To obtain a level of comfort with the selected discount rate, sensitivity testing was performed. The fourth category is the applicability of the factors to other asset classes, and the Academy recommends the use of the corporate bond factors for all fixed income assets, which is the current practice. Ms. Bennett explained that the C-1 category is directly tied to the bond’s rating from a nationally recognized statistical rating organization (NRSRO). Based on discussion with NRSROs, they use a global ratings process in which adjustments are made to the base factors such that an equivalent rating is produced whether it is a corporate bond, a public bond or a private bond. Therefore, the appropriate adjustments are already reflected on the NRSRO rating, and the Academy did not feel any further risk adjustments were necessary. © 2016 National Association of Insurance Commissioners

2

8

Attachment B Attachment XX Capital Adequacy (E) Task Force 12/xx/16 Mr. Fry said these four topics will be discussed further on the Working Group’s next conference call. Jerry Holman (Academy) provided an interim report on the Academy’s ongoing review of the portfolio adjustments currently included in the life RBC formula. He said the purpose of the portfolio adjustment factors is to scale a company’s average portfolio base factor such that the 92nd percentile confidence level for individual securities is converted to a 96th percentile target level for a given company’s portfolio. The Academy is following a similar approach to what is currently used in the life RBC formula that is based on the number of issuers. The factor decreases as the number of issuers increases. This approach recognizes that as the number of issuers increases, the diversification of the portfolio increases, and accordingly, the relative risk in the portfolio decreases. He said the Academy has expanded on the original approach and has modeled the portfolio for every company holding bonds subject to the C-1 factor down to as low as portfolios with three issuers, which is just under 700 companies. The current review has been expanded to also consider the size of the issuer held, as opposed to just the number of issuers. The Academy is still considering the cost of a more complex adjustment versus the benefit of greater precision. Mr. Holman said that the Academy has found that the current formula does not accurately capture risk, so it is important to update both the base factors and the adjustment factor at the same time. He said the Academy has found that the adjustment related to the top 10 issuers should be retained. Mr. Holman said a formal report with recommendations will be sent to the Working Group in the near future. Mr. Fry said the Academy’s presentation most likely will occur at the Fall National Meeting. Having no further business, the Investment Risk-Based Capital (E) Working Group adjourned. W:\QA\RBC\IRBC\Conference Calls and Meetings\2016\10-20 Call\Oct 20 IRBC CC Minutes.docx


3

9

10

Attachment C

Portfolio Adjustments to the C1 Factors for Corporate Bonds Presentation to the NAIC Investment Risk-based Capital Working Group December 11, 2016 Nancy Bennett, MAAA, FSA, CERA Rich Owens, MAAA, FSA, CFA American Academy of Actuaries C1 Work Group

Copyright © 2016 by the American Academy of Actuaries. All Rights Reserved.

11

Agenda

Attachment C



Purpose of Portfolio Adjustments



Current Portfolio Adjustments



Conceptual Methodology for Developing Adjustments



Considerations for IRBC



Next Steps for C1WG


2

12

Background on the Portfolio Adjustment Factor 

Ensure that the statistical safety level for the C1 component is met.    



Attachment C

Base C1 factors are set at the 92nd percentile over a 10-year time horizon for individual bonds Statistical safety target for the C1 component for an individual insurer’s bond portfolio is the 96th percentile over a 10-year time horizon The goal of the portfolio adjustment (PA) is to scale the base factors up or down, such that the 96th percentile target is achieved The adjustment for the 10 largest holdings reflects concentration risk and has no bearing on the statistical safety level; the top 10 adjustment is unrelated to the PA

In practice, for an individual insurer,  

More issuers in the bond portfolio narrow the loss distribution, justifying a lower C1 requirement A wider distribution of the issuer amount widens the loss distribution, justifying a higher C1 requirement


3

13

Current PA Factor for Portfolio Size Issuers

Attachment C

Factor

Up to

50

2.5

Next

50

1.3

Next

300

1.0

Over

400

0.9

Current Size Adjustment Factor 3.00 2.50 2.00 1.50 1.00 0.50

• In current LRBC formula, “size 0.00 0 adjustment factor” is the PA factor • Apply as sliding scale to derive weighted average factor • Example 500 Issuers: 1.16 = (50*2.5+50*1.3+300*1.0+100*0.9)/500 • Wtg average size adjustment factor times average base factor is portfolio C1 Copyright © 2016 by the American Academy of Actuaries. All Rights Reserved.

4

500

1000

1500

2000

Number of Issuers

14

Observations: Current Portfolio Adjustment

Attachment C



Only based on the number of issuers within a portfolio



Overstates the diversification benefit for small portfolios and understates for large portfolios



Therefore, C1 bond requirements are understated for small portfolios and overstated for large portfolios


5

15

Updating the PA: C1WG Working Construct

Attachment C



Update the portfolio factors for number of issuers (PA Alternative 1)



Evaluate a new PA measure designed to capture the variation in invested amount by issuer in addition to number of issuers (PA Alternative 2) (details to follow)



Meanwhile, retain the “top 10” adjustment to account for concentration risk


6

16

Portfolio Adjustment Factors: Overview of Methodology

Attachment C



Followed a similar approach to the development of the current “Size Adjustment Factor” to update the PA



Calculated the C1 component for 677 insurers’ bond portfolios from the NAIC data



Set the Target C1 as the C1 amount at the 96th percentile for each of the 677 bond portfolios   

Expanded original work that modeled a limited number of portfolios to consider every life company portfolio Based updated adjustment factors on data from 677 companies Used same company and issuer data used in base factor development


7

17

Portfolio Adjustment Factors: Overview of Methodology (cont.)

Attachment C



Determine a methodology to adjust the average base factors (up or down) creating an Adjusted C1 that matches the Target C1



Methodology is evaluated by the fit achieved: how close is the Adjusted C1 to the C1 target across all insurers? 

Ideally the fit is perfect and the Adjusted C1% for each company equals the Target C1% for that company (i.e., the difference is zero)  Best fit minimizes error, defined as the average of the differences between the Adjusted C1% to the Target C1% 

The PA factor scales the base factors, such that the 96th percentile target is achieved and has better fit by company


8

18

PA Alternative One: Number of Issuers Only

Up to Next Next Next Over

Issuers

7 33 160 550 750

Updated Number of Issuer Factor

Factor 7.55 3.70 0.85 0.80 0.75

8.00 7.00 6.00 5.00 4.00 3.00 2.00 1.00 0.00

• Apply as per sliding scale of current formula • Example 500 Issuers Factor = 1.10 • Factor times average base factor is portfolio C1


Attachment C

0

500

1000

1500

2000

Number of Issuers

9

19

Current PA vs. Updated PA Alternative 1 (number of issuers only)

Attachment C



PA Alternative 1 corrects for bias of less than target C1 for portfolios with less than 50 issuers and bias of more than target C1 for portfolios with high number of issuers.



Companies with greater issuer amount variation, as measured by Coefficient of Variation (CV) are more likely to be target outliers relative to the target for C1.

Difference in $millions to Target C1 150.0 100.0 50.0 .0 -50.0

0

500

1000

1500

2000

2500

-100.0 -150.0

Number of Portfolio Issuers Current PA

PA Alternative 1


10

20

PA Alternative Two: # Issuers and Issuer Amount Distribution

Attachment C

PA factor = Average Issuers Factor + CV Factor

Up to Next Next Next Over

Issuers & CV Number of Issuers Factor 7 7.350 33 2.850 160 0.325 500 0.130 700 0

Plus

Issuer and CV PA Factor 9.00 8.00 7.00 6.00 5.00 4.00 3.00 2.00 1.00 0.00 0

CV More Than

500

1000

1500

2000

Number of Issuers CV Low

Up To Factor 0.00 0.45 0 0.45 0.65 0.400 0.65 0.85 0.550 0.85 1.20 0.650 1.20 1.55 0.750 1.55 2.00 0.800 2.00 3.00 0.850 3.00 1.500


11

CV Mid

CV High

21

PA Alternative Two: Number of Issuers Plus CV



Alt 2 tightens the range of difference for companies with under 1300 issuers



Results mixed 1500-2000 issuers, some closer to 0, some change from minus to plus, other from plus to minus



Over 2000 issuers, two of three results better, one switches sign

Difference in $millions to Target C1 150.0 100.0 50.0 .0 -50.0

0

500

1000

1500

2000

2500

-100.0 -150.0

Number of Portfolio Issuers PA Alternative 1

PA Alternative 2


12

Attachment C

22

Attachment C

Calculating PA Alternative Two 

Portfolio Unadjusted C1 = 1.20%, Target C1 = 1.07%



Portfolio has 843 Issuers, 



PA based on the # issuers is 0.31 (from table for PA2)

Portfolio has CV of 0.61 

PA based on CV has CV Factor = 0.40



Adjustment factor = Average Issuers Factor + CV Factor = 0.31 + 0.40 = 0.71



Adjusted C1 = 1.20% * 0.71 = 0.86%



Error = (Target – Adjusted) = 0.22%


13

23

PA Alternative One vs. Two 

Developed two variations of potential PAs by minimizing overall differences of C1 target to individual results



Ideal average differences error is zero Average Differences Error

Current PA

Alt PA1

Alt PA2

0.25%

0.10%

0.07%

1.4

-0.1

0

C1$ bil – Target C1$ bil


14

Attachment C

24

Next Steps 

Attachment C

Get IRBC Feedback  

Number of issuers only Number of issuers and CV



Finalize model and documentation of PAs



Recommend to IRBC


15

25

Questions

Attachment C

For more information, please contact: Nancy Bennett, Academy Senior Life Fellow [email protected] Amanda Darlington, Academy Life Policy Analyst [email protected] (202) 223-8196


16

26

Appendix

Attachment C

Details on Coefficient of Variation (CV)


17

27

Issuer Amount Distribution

Attachment C



Consider Risk of 2 portfolios of $100 million



Port 1: 10 issuers of $10 million each



Port 2: 1 issuer of $91 million, 9 issuers of $1 million



Is the risk the same?


18

28

Issuer Amount Distribution 

Can be measured by the Coefficient of Variation (CV)  





Attachment C

The CV is a measure of spread that describes the amount of variability relative to the mean. The CV is an alternative to standard deviation and a better statistical measure when comparing distributions of different sizes. CV equals the standard deviation divided by average of issuer amounts held by a company

Data is anticipated to be available from identical data source used to calculate top ten concentration factor for bonds


19

29

30

Attachment D

John Bruins Vice President & Senior Actuary 202.624.2169 t [email protected]

October 10, 2016 Mr. Kevin Fry Chair, Inv. RBC Working Group Re:

RBC C-1 Factors for Bonds Alternative Proposal

Dear Kevin: In August 2015, new C-1 factors for bonds and similar assets to be used in the NAIC life RBC calculation were proposed by the American Academy of Actuaries (“Academy’). These factors reflect a dramatic change in the capital charges, resulting in a shift of incentives from investment grade to below investment grade. After extensive analysis, we are convinced that this shift is not simply due to updated experience, but additionally is due to modeling choices which, in combination, do not appropriately reflect the underlying risks. The ACLI1 is recommending an alternate set of C-1 factors to be considered for adoption by the NAIC as show in Appendix A. We believe these alternative factors better reflect the impact of the underlying experience, and are consistent with the guidelines and parameters articulated for such capital factors in the life RBC formula. The purpose of this letter is to provide some background rationale for the development of these revised factors, as well as documentation about how they were developed. Adopting this ACLI proposal has several benefits that will be discussed later in the letter, namely: 1. It is completely transparent. The data sources are documented and publicly available. The model has been shared with the NAIC Structured Securities Group for review and analysis for the benefit of the regulators; 2. It fixes several model assumptions used by the Academy that do not accurately reflect portfolio credit risk; 1

The American Council of Life Insurers (ACLI) is a Washington, D.C.-based trade association with approximately 280 member companies operating in the United States and abroad. ACLI advocates in state, federal, and international forums for public policy that supports the industry marketplace and the 75 million American families that rely on life insurers’ products for financial and retirement security. ACLI members offer life insurance, annuities, retirement plans, long-term care and disability income insurance, and reinsurance, representing 95 percent of industry assets, 92 percent of life insurance premiums, and 97% of annuity considerations in the United States. Learn more at www.acli.com.

American Council of Life Insurers 101 Constitution Avenue, NW, Washington, DC 20001-2133 www.acli.com

31

Attachment D

2 3. It provides a basis for setting risk appropriate factors for all major fixed coupon assets classes (i.e. corporates, municipal bonds, privates, and sovereigns), which industry is supportive of; and 4. It produces results that are consistent with similar studies done by other portfolio risk analysts (i.e. S&P/Moody’s/Industry) On September 29, 2015 ACLI submitted comments on the proposal from the Academy that was exposed by the Inv. RBC working group in August, 2015 to revise the RBC factors for Bonds. That letter outlined a series of concerns that we had identified based on initial review of the documentation provided by the Academy. Following submission of the letter, we have had subsequent meetings with the Academy to better understand the assumptions, the modeling approach and the decisions behind their recommendation as well as to clarify our initial concerns. Those discussions proved insightful, but have not resulted in any changes to the recommendation. Using the documentation provided and information learned from the direct discussions with the Academy, ACLI members undertook a more indepth analysis of the Academy model to understand the sources of the significant change in many of the factors and counterintuitive nature of those changes relative to other analysis of credit risk available in the marketplace. This analysis raised a number of questions about many of the assumptions and modeling choices made and the potential for them to either materially overstate or understate the credit risk. Given the complexity of interactions among different modeling segments and the sheer number of factor and modeling decision points, we thought it most useful to assess materiality by re-modeling using appropriate methods and observing the results. As the Academy did not share their model directly, investment professionals in the industry subsequently built a model which closely replicates the published Academy factors. After replicating the results, we developed changes to assumptions and modeling approaches to address material concerns that came to light during the initial evaluation process. Some of the revised assumptions, such as using issuer-weighted recovery data to parameterize issuer recovery in the model, fit easily into the framework created by the Academy. Other updates, such as modeling the portfolio of assets as a portfolio instead of separately by rating grade, require more significant overhauls to the model. While no model is ever entirely accurate, the ACLI has made updates to the proposed Academy modeling that much more closely align the individual pieces to the reality observed in credit markets since the 1980s. Importantly, we also utilized only publicly available data in place of the unpublished source of information for recovery information so that the modeling could be freely shareable and completely transparent. The following outlines the key changes in assumptions and approach from that used by the Academy which are documented more fully in Appendix B.:

32

Attachment D

3 1. Portfolio modeling approach – The ACLI model takes a full representative portfolio differentiated by every modeled asset class and rating, then simulates the entire portfolio in each path, and finally attributes losses to ratings based on rating losses in the appropriate CTE90 paths, rather than modeling each rating independently with NAIC 1-rated representative holdings at the 92nd percentile. 2. Economic states – Rather than using discrete annual economic states, the model relies on a credit default factor that can effectively replicate the cyclical nature of credit. 3. Default assumptions – ACLI and AAA used the same underlying source of data, but different techniques for smoothing. The ACLI relies on Moody’s published process for smoothed default rates. 4. Recovery assumptions – ACLI and AAA used the same default and recovery database. The ACLI approach uses issuer-weighted rather than bond-weighted recovery data, which is both more conservative and more appropriate. 5. Reinvestment following default – Rather than allow the reinvestment of full principal after default, the ACLI model assumes only the recovered amount is reinvested. 6. Offset for reserve provisions – The ACLI model uses rating-specific CTE70 default losses from the model as the risk premium or spread, consistent with that assumed in reserve provisions and different from the Academy’s use of average default losses. 7. Other Asset Classes – The ACLI model differentiates inputs between asset classes by using distinct recovery assumptions and implied default rates maintaining the same expected loss throughout by rating. Based on the foregoing, the ACLI recommends the C1 factors shown in Table 1 as appropriate factors meeting the guidelines and parameters that have been articulated for RBC factors for bonds in the Life RBC formula. We will be happy to discuss these recommendations with the Working Group, the Academy, and the NAIC Structured Securities Group.

s/ John Bruins cc

Julie Garber, NAIC Nancy Bennett, American Academy of Actuaries

33

Attachment D

4

Appendix A: Comparison of Pre-Tax C1 Bond Factors Table of Pre-Tax Factor Comparison Rating Category Current Proposed

NAIC

Academy

Rating

Current

Proposed

0.40% 0.40% 0.40% 0.40% 0.40% 0.40% 0.40% 1.30% 1.30% 1.30% 4.60% 4.60% 4.60% 10.00% 10.00% 10.00% 23.00% 23.00% 23.00% 30.00%

0.34% 0.34% 0.72% 0.72% 1.16% 1.16% 1.16% 1.49% 1.68% 2.01% 3.55% 4.39% 5.62% 5.99% 7.86% 10.31% 17.31% 17.31% 17.31% 30.00%

1 1 1 1 1 1 1 2 2 2 3 3 3 4 4 4 5 5

1-A 1-B 1-C 1-D 1-E 1-F 1-G 2-A 2-B 2-C 3-A 3-B 3-C 4-A 4-B 4-C 5-A 5-B

Aaa Aa1 Aa2 Aa3 A1 A2 A3 Baa1 Baa2 Baa3 Ba1 Ba2 Ba3 B1 B2 B3 Caa1 Caa2

5 6

5-C 6

Caa3 DFLT

ACLI Public US Government Privates Corporate Municipals Related 0.33% 0.18% 0.22% 0.27% 0.35% 0.19% 0.23% 0.29% 0.37% 0.20% 0.25% 0.30% 0.41% 0.22% 0.27% 0.33% 0.47% 0.25% 0.31% 0.38% 0.56% 0.30% 0.37% 0.45% 0.71% 0.38% 0.47% 0.58% 0.89% 0.50% 0.61% 0.73% 1.13% 0.65% 0.78% 0.94% 1.77% 1.03% 1.23% 1.48% 2.24% 3.02% 4.15% 5.47% 7.40% Same as Public Corporate 10.27% 15.19% 22.55% 27.00% 30.00%

34

Attachment D

5

Chart of Pre-Tax Factor Comparison

35

Attachment D

6

Appendix B: Detailed Description of ACLI Model Updates Portfolio Modeling Approach C1 bond RBC is intended to represent the minimum capital needed for an insurer to cover the credit risk inherent in its bond portfolio. As such, the factors by credit rating should be determined from a model that does not ignore the context of the insurer bond portfolio: representative holdings must represent the differences in holdings size and number by credit rating and the simulated “tails” need to be those paths where the portfolio performs poorly. By contrast, the Academy model simulates each credit rating completely independently. Each of the rating categories (initially 19 non-default categories from Aaa to Caa3, grouped to 13 categories) is assigned to holdings of approximately 400 issuers, a number roughly representing the number of NAIC1rated issuers held in a typical insurer’s portfolio. This approach misrepresents portfolio credit risk in several ways:  Overstates risk: the benefit of diversification among different credit ratings is lost. The main driver of higher capital factors in any simulated path is whether many large issuers default together. All rating categories would not have their worst default performance at the same time, especially when considering the distribution of issuer size. The use of the 92nd percentile for each rating attempts to adjust for this but can only do so approximately.  Understates risk: none of the credit ratings, when using either as 19 or 13 risk categories, has anywhere near holdings of 400 issuers. The typical insurer portfolios used to develop the assumption of 400 NAIC1 issuers have approximately 1000 total issuers – using 19 rating categories of 400 issuers implies 7.5 times more issuers than ae actually held by companies. The less issuers there are, the more severe the outcomes in the tail of the distribution of results.  Misstates risk: o The better the credit rating, the more willing insurers are to have larger holdings in a single issuer. This effect is ignored by using a single issuer size distribution across all ratings. o The relative importance of different credit ratings to insurers’ portfolios is lost. A and Baa credit ratings compose the majority of insurers’ holdings of both issuers and dollars. The modified model developed by the ACLI addresses all of these concerns. The portfolio is simulated across all credit ratings together with capital determined by the worst 10% of paths for the entirety of the portfolio, in line with the NAIC’s CTE90 standard for RBC. The updated representative holdings account for the differences in aggregate holding size of each credit rating in addition to the different distributions of issuer size within each credit rating. Additionally, the representative holdings are derived from the same year-end 2011 data used by the Academy but are now derived from a much larger sample of insurers’ portfolios: 177 versus 24 for the Academy.

36

Attachment D

7 By not holistically modeling as a portfolio, the Academy approach overstates the risk of higher credit ratings relative to lower credit ratings and, for the same confidence level, overstates the level of capital required.

Economic States While US recessions and global credit cycles have historically had overlaps, the two do not exhibit a perfect correlation; recessions are also not all equivalent in terms of severity. Nonetheless, to generate variation in credit inputs in its model, the Academy linked annual credit input data with GDP-based categorizations of each year into recessions or expansions, including added effects for more than consecutive year of either state. While credit certainly tends to deteriorate during recessions, a simpler and more accurate approach would be to reproduce the behavior of credit observed throughout the historical data without the added constraint of tying to GDP. Credit moves in long cycles, which has important implications for the accuracy of the credit modeling:  Defaults are usually very low for many years before subsequently rising for more than one or even two years in a row when a credit downturn occurs. By not imposing several consecutive years of elevated defaults, the Academy model understates the impact of normal downturns.  Even in the most severe downturns such as the Great Depression, there were only 5 years of 10 that would be considered recessionary. In the Academy model, more than a handful of extreme simulated paths can have as many as 7 years of 10 as recessions. This suggests that the Academy model overstates the impact of severe downturns.  Issuers with lower credit ratings are more sensitive to elongated periods of stress due to their lower margin for error. This manifests itself in the modeling through their higher default rates.

The modified model from the ACLI reinterprets the same Moody’s default study data used by the Academy to create an aggregate credit default factor through time. It is shown above along with the Academy’s recession, or contraction, periods. This credit default factor is defined as the yearly average of each available credit rating’s realized default rate, normalized by that rating’s idealized default rate

37

Attachment D

8 (see next section) and historical standard deviation as well as log-transformed. This simple credit default factor shows statistically significant levels of cyclicality, which is then simulated forward using an autoregressive process that imitates this cyclicality. By not capturing the cyclicality of credit, on balance the Academy model understates the level of capital required and overstates the risk of higher credit ratings relative to lower credit ratings.

Default Assumptions Since default is a quite rare event, historical probability of default (PD) data is noisy. PDs must be smoothed in order to produce sensible results, and this process demands considerable judgment. The Academy’s PDs have term structures that are surprisingly volatile – in many cases much more than the underlying data – which are the result of smoothing cumulative default rates instead of marginal default rates. The raw marginal default rates from Moody’s Corporate Default and Recovery Rates 1920-2012 and the Academy’s smoothed version of these are shown below on the left for A1 through Baa3 ratings:

38

Attachment D

9 Instead of designing an improved smoothing methodology, the ACLI model uses smoothed inputs directly from Moody’s2. When rating structured products, Moody’s applies a set of idealized expected losses (ELs) by instrument rating to the hundreds of pieces of debt in the collateral. In combination with a long-run assumption on senior unsecured loss given default (LGD) from the Moody’s default study, a set of idealized PDs is thus implied from the identity EL = PD x LGD. These PDs are shown above on the right side for comparison purposes. Swapping the Academy’s smoothed PDs for those implied by the Moody’s idealized inputs indicates that the Academy approach overstates the risk of higher credit ratings relative to lower credit ratings.

Recovery Assumptions The Academy model uses loss given default (LGD) data provided by S&P, which has some counterintuitive features. In particular, the average LGD of 53% is quite low for senior unsecured bonds and LGD has historically been better in recessionary time periods rather than worse. The ACLI has closely replicated the data the Academy used through the Moody’s Default & Recovery Database, and has determined that the root cause of the abnormalities is the use of bond-weighted instead of issuerweighted data. This allows a handful of defaulted companies with a large number of issues to skew the dataset. Additionally, when parametrizing a model that tracks issuers, it is more valid to use data collected at the issuer level. The summary table shown below on the left illustrates how issuerweighting LGDs leads to much more reasonable inputs.

The modified model from the ACLI uses issuer-weighted LGD data, however instead of sourcing public corporate data from a proprietary source it relies on publicly available Moody’s studies3. The issuer2 3

. Moody's Global Approach to Rating Collateralized Loan Obligations", Appendix 1, September 2015 Moody's studies include each “Corporate Default and Recovery Rate” study since 2004 and the 2003 study “Recovery Rates on Defaulted Corporate Bonds and Preferred Stocks, 1982-2003”

39

Attachment D

10 weighted average and standard deviation of LGDs by year is available from these sources, which the ACLI model employs to estimate a relationship with the credit default factor mentioned previously. As one would expect, and as can be seen on the scatterplot above, the relationship between average LGD and the credit default factor is quite strong for both public corporates and private corporates4. This highlights the appropriateness of both the LGD data as well as the credit default factor, and suggests that the ACLI model will be able to mimic all parts of the distribution well. By contrast, the Academy approach understates the level of capital required and overstates the risk of higher credit ratings relative to lower credit ratings by using LGDs that are both too low and that do not get worse during stressed times.

Reinvestment Assumption In the Academy model, defaults are followed by a reinvestment of the par value of the defaulted bond rather than the recovered amount. This implies a capital infusion. In contrast, the ACLI model assumes reinvestment of the recovered amounts. While this only has a negligible impact for all but the riskiest credit ratings because multiple defaults for a particular issuer over 10 years are highly unlikely, it nonetheless is clearly more realistic. In addition, this is an easy assumption to implement so it features in the ACLI’s upgraded model.

Offset for Reserve Provisions In the ACLI letter of Sept. 29, 2015, we questioned the offset for reserve provision. While stating that reserves are held at a CTE70 or one standard deviation above the mean, the Academy modeling used mean expected losses as an offset. The old reserve method did not have an explicit provision for credit loss, but rather is set conservatively overall. Principle-Based Reserves are more explicit, and specifically use default costs that are set at CTE70. For the sake of consistency with reserves, the ACLI model uses simulated CTE70 discounted default costs as the risk offset for each rating. As can be seen on the chart below, these offsets are higher across the board than that used by the Academy but still much lower than the full average spread observed in Barclays data since the early 1990s. It is also very important to note that the Academy Risk Premium incorporates the Academy’s low LGD estimate, which causes the risk offset to be understated.

4

Private corporate yearly issuer-weighted data from Society of Actuaries 2006 study “1986-2002 Credit Risk Loss Experience Study: Private Placement Bonds”

40

Attachment D

11

By not explicitly accounting for the CTE70 level implicit in reserves, the Academy model double-counts losses between capital and reserves. This effect is larger for higher credit ratings, which causes the Academy model to overstate the risk of higher credit ratings relative to lower credit ratings.

Other Asset Classes The Academy proposal recommends that capital factors for any given credit rating be the same across all asset classes. The rationale is that rating agencies use a global ratings process that is associated with an expected loss that is identical across the asset classes, and therefore capital factors should be identical as well. Capital is the amount of stressed loss in excess of the average loss. Expected loss is this average loss. These two values are not always proportional. The composition of the expected loss is in fact critical in determining capital. Referring back to the identity EL = PD x LGD, two assets can have the same ELs but different PDs and LGDs. Models and intuition indicate that the asset with higher PD and lower LGD will need less capital since less dollars will be needed to cover the losses in any particular path.

41

Attachment D

12

Asset Class Private Corporates US Municipals Other GovernmentRelated Public Corporates

Average LGD 29%5 40%6 50%7

63%8

The upgraded ACLI model assumes that all asset classes have the same expected loss by rating, consistent with the concept that rating agencies use an equivalent “global rating scale”. As a reminder, these are assumed to be the idealized Moody’s expected losses discussed earlier. Differentiation shows up because of the different historically observed recovery rates by asset class with private corporates having the best LGD at about 30%, or less than half of that of public corporates. By using the recovery rates observed for these asset types, we could develop the implied expected default rates, which could then be used to model stress environments. The resulting factors are included in the proposed factors of Table 1. Given the lack of credible data, revised factors are recommended only for the investment grades, but non-investment grade bonds are recommended to use the same factors as for the public corporate bonds. By not differentiating by asset class, the Academy model overstates the level of capital required, especially so for assets at high credit ratings.

5

Derived from issuer-weighted LGDs from two Society of Actuaries studies: April 2006 – “1986-2002 Credit Risk Loss Experience Study: Private Placement Bonds” , January 2016 – “2003-12 Credit Risk Loss Experience Study: Private Placement Bonds” 6 Moody’s “US Municipal Bond Defaults and Recoveries, 1970 – 2014”, July 2015 7 Moody’s “Sovereign Default and Recovery Rates, 1983 – 2014”, April 2015 8 Moody’s “Corporate Default and Recovery Rates, 1920 – 2015”, February 2016

42

Attachment D

13

Appendix C: Illustrative Model Step-Through Analysis In order to help illustrate the impact of some of the ACLI’s updated assumptions, we walk through the cumulative impact of certain changes to the Academy model. Since the Academy did not share its model, we first present a comparison of the Academy’s published modeled charges with the factors generated by the ACLI’s rebuilt version of the Academy model. As can be seen in the chart below, the ACLI rebuilt model produces factors that are nearly identical to that from the Academy with the exception of Caa3. Note that this chart as well as all others in this appendix are on a logarithmic scale, meaning that two parallel lines have a constant percent difference between all points.

43

Attachment D

14

Step I: Updating Recovery and Reinvestment Assumptions The first step we take is to update the recovery assumptions everywhere in the Academy model to use the issuer-weighted version instead of bond-weighted version of the LGD data. This change – moving from the blue line to the gold line on the chart below – increases factors across all credit ratings, and increases factors in lower credit ratings than higher ratings on a percentage basis. As described earlier, both the average LGD as well as the LGD in recessions relative to non-recessions is getting worse with this updated assumption. Note that these exact inputs are not used in the ACLI’s updated model, as that model uses Moody’s published issuer-weighted LGD information in the form of averages and standard deviations instead of LGDs bucketed into 10% bands. We also take another half-step here to illustrate the impact of reinvesting only the recovered principal after default instead of the full par amount. Within the context of the Academy model, this does not have much of an impact as it only visibly impacts B and Caa (see orange line on chart relative to gold line). The intuition here is that for the lower risk credit ratings, low default rates mean it is highly unlikely an issuer will default multiple times within 10 years so the assumption has no effect. However, in B and Caa, reinvesting full par means there is more principal that can be lost again in a future default. In fact, at higher percentiles these losses can exceed 100%.

44

Attachment D

15

Step II: Updating Probabilities of Default In the next step we update the Academy’s smoothed probabilities of default (PDs) to be those implied by Moody’s idealized Expected Losses (ELs) divided by an overall public corporate issuer-weighted senior unsecured LGD assumption of roughly 63%. As with step 1, we also update risk premiums or spreads to reflect the changed assumption. Note however that we do not change the calibrated economic state PD multipliers. As seen in the chart below moving from blue to orange, the net effect of this change is to substantially steepen the slope of charges.

45

Attachment D

16

Step III: Raw Output of ACLI Model Lastly, we present the raw output of the full ACLI model. Many of the updates here cannot be sensibly disaggregated from one another, so we do not attempt to do so. The main changes are modeling as a portfolio – including using separate asset classes – and having a credit default factor at the heart of the simulation. Note that for the non-public corporate asset classes, there is often not enough holdings or notched information at certain rating categories so the holdings are grouped together. For privates, we only know current NAIC ratings, so these are grouped together as the simple average of the notched rating inputs (i.e. Baa Private Expected Loss is average of Expected Loss of Baa1, Baa2, and Baa3). We also assume that all NAIC 1 rated privates are A-rated. For US Municipals and Other GovernmentRelated categories, almost all holdings are NAIC 1 and are difficult to find notched rating information so we assume the average of Aaa through A3 for our average inputs.

46

Attachment D

17

Appendix D: Modeling at Shorter Time Horizons During the September 2016 RBC C1 Working Group call, a comment was made that it is not possible to run the Academy model at shorter time horizons without much additional parameterization work. To our understanding this is not correct. We present below the results from the ACLI’s rebuilt version of the Academy model at shorter time horizons, and the resulting changes in factors largely behave as one might expect given such as change. These additional time horizons can also be easily calculated in the ACLI proposed model.

47

Materials - Investment Risk-Based Capital (E) Working Group

Recommend Documents