policyholder behavior experience data and modeling timothy ... · presentation disclaimer....
TRANSCRIPT
Equity-Based Insurance Guarantees Conference Nov. 11-12, 2019
Chicago, IL
Policyholder Behavior Experience Data
and Modeling
Timothy Paris
SOA Antitrust Compliance Guidelines SOA Presentation Disclaimer
Sponsored by
SOA Equity-Based Insurance Guarantees Conference
Policyholder Behavior Experience Data and ModelingSession 2B
TIMOTHY PARIS, FSA, MAAA
RUARK CONSULTING, LLC
November 11, 2019
1:30 - 3:00pm Central
SOA
Antit
rust
Com
plia
nce
Gui
delin
esAc
tive
part
icip
atio
n in
the
Soci
ety
of A
ctua
ries
is a
n im
port
ant a
spec
t of m
embe
rshi
p. W
hile
the
posi
tive
cont
ribut
ions
of p
rofe
ssio
nal s
ocie
ties
and
asso
ciat
ions
are
wel
l-rec
ogni
zed
and
enco
urag
ed, a
ssoc
iatio
n ac
tiviti
es a
re v
ulne
rabl
e to
clo
se a
ntitr
ust s
crut
iny.
By
thei
r ver
y na
ture
, ass
ocia
tions
brin
g to
geth
er in
dust
ry c
ompe
titor
s an
d ot
her m
arke
t pa
rtic
ipan
ts.
The
Uni
ted
Stat
es a
ntitr
ust l
aws
aim
to p
rote
ct c
onsu
mer
s by
pre
serv
ing
the
free
eco
nom
y an
d pr
ohib
iting
ant
i-com
petit
ive
busi
ness
pra
ctic
es; t
hey
prom
ote
com
petit
ion.
The
re
are
both
stat
e an
d fe
dera
l ant
itrus
t law
s, a
lthou
gh st
ate
antit
rust
law
s cl
osel
y fo
llow
fede
ral l
aw.
The
Sher
man
Act
, is t
hepr
imar
y U
.S. a
ntitr
ust l
aw p
erta
inin
g to
ass
ocia
tion
activ
ities
. T
he S
herm
an A
ct p
rohi
bits
eve
ry c
ontr
act,
com
bina
tion
or c
onsp
iracy
that
pla
ces
an u
nrea
sona
ble
rest
rain
t on
trad
e. T
here
are
, how
ever
, som
e ac
tiviti
es th
at a
re il
lega
l un
der a
ll ci
rcum
stan
ces,
suc
h as
pric
e fix
ing,
mar
ket a
lloca
tion
and
collu
sive
bid
ding
.
Ther
e is
no
safe
har
bor u
nder
the
antit
rust
law
for p
rofe
ssio
nal a
ssoc
iatio
n ac
tiviti
es.
Ther
efor
e, a
ssoc
iatio
n m
eetin
g pa
rtic
ipan
ts sh
ould
refr
ain
from
dis
cuss
ing
any
activ
ity th
at
coul
d po
tent
ially
be
cons
true
d as
hav
ing
an a
nti-c
ompe
titiv
e ef
fect
. Dis
cuss
ions
rela
ting
to p
rodu
ct o
r ser
vice
pric
ing,
mar
ket a
lloca
tions
, mem
bers
hip
rest
rictio
ns, p
rodu
ct
stan
dard
izat
ion
or o
ther
con
ditio
ns o
n tr
ade
coul
d ar
guab
ly b
e pe
rcei
ved
as a
rest
rain
t on
trad
e an
d m
ay e
xpos
e th
e SO
A an
d its
mem
bers
to a
ntitr
ust e
nfor
cem
ent p
roce
dure
s.
Whi
le p
artic
ipat
ing
in a
ll SO
A in
per
son
mee
tings
, web
inar
s, te
leco
nfer
ence
s or
sid
e di
scus
sion
s, y
ou s
houl
d av
oid
disc
ussi
ngco
mpe
titiv
ely
sens
itive
info
rmat
ion
with
com
petit
ors
and
follo
w th
ese
guid
elin
es:
•-D
o no
tdis
cuss
pric
es fo
r ser
vice
s or
pro
duct
s or a
nyth
ing
else
that
mig
ht a
ffect
pric
es
•-D
o no
tdis
cuss
wha
t you
or o
ther
ent
ities
pla
n to
do
in a
par
ticul
ar g
eogr
aphi
c or
pro
duct
mar
kets
or w
ith p
artic
ular
cus
tom
ers.
•-D
o no
tspe
ak o
n be
half
of th
e SO
A or
any
of i
ts c
omm
ittee
s un
less
spec
ifica
lly a
utho
rized
to d
o so
.
•-D
ole
ave
a m
eetin
g w
here
any
ant
icom
petit
ive
pric
ing
or m
arke
t allo
catio
n di
scus
sion
occ
urs.
•-D
oal
ert S
OA
staf
f and
/or l
egal
cou
nsel
to a
ny c
once
rnin
g di
scus
sion
s
•-D
oco
nsul
t with
lega
l cou
nsel
bef
ore
rais
ing
any
mat
ter o
r mak
ing
a st
atem
ent t
hat m
ay in
volv
e co
mpe
titiv
ely
sens
itive
info
rmat
ion.
Adhe
renc
e to
thes
e gu
idel
ines
invo
lves
not
onl
y av
oida
nce
of a
ntitr
ust v
iola
tions
, but
avo
idan
ce o
f beh
avio
r whi
ch m
ight
be
so c
onst
rued
. Th
ese
guid
elin
es o
nly
prov
ide
an o
verv
iew
of
pro
hibi
ted
activ
ities
. SO
A le
gal c
ouns
el re
view
s m
eetin
g ag
enda
and
mat
eria
ls a
s de
emed
app
ropr
iate
and
any
dis
cuss
ion
that
dep
arts
from
the
form
al a
gend
a sh
ould
be
scru
tiniz
ed c
aref
ully
. An
titru
st c
ompl
ianc
e is
eve
ryon
e’s
resp
onsi
bilit
y; h
owev
er, p
leas
e se
ek le
gal c
ouns
el if
you
hav
e an
yqu
estio
ns o
r con
cern
s.
2
Presentation Disclaimer
Presentations are intended for educational purposes only and do not replace independent professional judgment. Statements of fact and opinions expressed are those of the participants individually and, unless expressly stated to the contrary, are not the opinion or position of the
Society of Actuaries, its cosponsors or its committees. The Society of Actuaries does not endorse or approve, and assumes no responsibility for, the content, accuracy or completeness of the
information presented. Attendees should note that the sessions are audio-recorded and may be published in various media, including print, audio and video formats without further notice.
3
Background
4
Why is this important?
• Critical risk element for long-term EBIG-type products, and significant interactions with capital markets risks
• Complex and dynamic data
• Data sparsity, especially at the company level – a credibility problem
• Data is emerging in key areas
• Asset-side folks – demand that your liability-side colleagues demonstrate the robustness of models/assumptions that they provide you
5
Industry studies
Fixed-indexed annuity policyholder behaviorhttps://ruark.co/ruark-releases-2019-fixed-indexed-annuity-study/https://ruark.co/ruark-consulting-releases-2018-fixed-indexed-annuity-mortality-study/
Variable annuity policyholder behaviorhttps://ruark.co/ruark-releases-2019-variable-annuity-study-results/https://ruark.co/ruark-consulting-releases-variable-annuity-mortality-study-results/
6
VM-21 PBR for Variable Annuities
Public redline exposure draft as of April 30, 2019https://naic-cms.org/exposure-drafts
Section 10: Contract Holder Behavior AssumptionsShould examine many factors including cohorts, product features, distribution channels, option values, rationality, static vs dynamicRequired sensitivity testing, with margins inversely related to data credibilityUnless there is clear evidence to the contrary, should be no less conservative than past experience and efficiency should increase over timeWhere direct data is lacking, should look to similar data from other sources/companies
1
2
3
4
7
You and your data
8
0%
35%
7 ormore
6 5 4 3 2 1 0 -1 -2 -3 ormore
Years Remaining in Surrender Charge Period
2008
2016
2018
Your company-level data might indicate some key patterns in surrender behavior
9
GLWB
Surr
ende
r Rat
e
9
0%
30%
7 ormore
6 5 4 3 2 1 0 -1 -2 -3 ormoreYears Remaining in Surrender Charge Period
Surrender rates are lower with living benefits…
10
Surr
ende
r Rat
e
None
GLWB
Hybrid GMIB
10
0%
25%
7 ormore
6 5 4 3 2 1 0 -1 -2 -3 ormore
Surr
ende
r Rat
e
Years Remaining in Surrender Charge Period
GLWB - Withdrawal Behavior
11
…and even lower with income utilization
No prior WDs
Excess WDs
Less than or full WDs
11
…and when guarantees are more valuable
12
GLWB (nominal moneyness basis)
0%
25%
7 ormore
6 5 4 3 2 1 0 -1 -2 -3 ormore
Surr
ende
r Rat
e
Years Remaining in Surrender Charge PeriodITM 50+% ITM 25 - 50% ITM 5 - 25% ATM OTM
12
Dynamic sensitivity has also changed over the years
13
0%
35%
3Q 09 3Q 10 3Q 11 3Q 12 3Q 13 3Q 14 3Q 15 3Q 16 3Q 17 3Q 18
GLWB Shock Lapse
ATM <25% ITM 25%-50% ITM 50%-100% ITM
Surr
ende
r Rat
e
13
How you measure value matters, but company-level credibility is very limited
14
0%
25%
OTM 50+% OTM 25 -50%
OTM 5 -25%
ATM ITM 5 -25%
ITM 25 -50%
ITM 50 -100%
ITM 100%+
Surr
ende
r Rat
eGLWB
Shock -actuarial
Ultimate -actuarial
Shock -nominal
Ultimate -nominal
14
Largest and smallest contracts behave differently
15
0%
20%
7 ormore
6 5 4 3 2 1 0 -1 -2 -3 ormore
Surr
ende
r Rat
e
Years Remaining in Surrender Charge Periodunder 50,000 50,000-100,000 100,000-250,000
250,000-500,000 500,000-1,000,000 >=1,000,000
15
Withdrawals vary by age and tax status
16
QualifiedNon-Qualified
0%
100%
<50 50-59 60-64 65-69 70-79 80+
Freq
uenc
y
Attained Age
GLWB
16
Withdrawal behavior is becoming more efficient
17
0%
50%
Q12008
Q12009
Q12010
Q12011
Q12012
Q12013
Q12014
Q12015
Q12016
Q12017
Q12018
Freq
uenc
y
GLWB
LT Full WDs Full WDs Excess WDs
17
0%
10%
<50 50-59 60-64 65-69 70-79 80-LEA Last Eligible
Annu
itiza
tion
Rate
Hybrid GMIB annuitization rates are low, but company-level credibility is very limited
1818
2012 IAM does not fit VA mortality experience very well
19
0%
150%
0-49 50-54 55-59 60-64 65-69 70-74 75-79 80-84 85-89 90+
% o
f Tab
le
Actual vs. 2012 IAM - Projection G2
Male Count Male Amount Female Count Female Amount Base
19
Evidence of anti-selection for death benefit guarantees
20
0%
200%
1 2 3 4 5 6 7 8 9 10 11+
% o
f Tab
le
Duration
Actual vs. 2012 IAM-G2
LB No LB
20
Results vary over time and between companies
• Each company’s size affects quality of analytical insights and volatility of their own results – a credibility problem
• Composition differences
• Idiosyncratic differences – product features, distribution, closed blocks, etc
• Using only your data, it is very difficult to identify the signal from the noise
21
Building models with your data
22
Modeling and assumptions
• Measuring goodness-of-fit for candidate models
• Testing predictive power on out-of-sample data
• Art + science: choosing, communicating, and ongoing recalibration
23
Goodness of Fit
Predictive Power
24
0
1,000
2,000
3,000
4,000
0 1 2 3 4 5 6 7 8 9 10
Baye
sian
Info
rmat
ion
Crite
rion
(BIC
)
Number of Factors
25
0% 2% 4% 6% 8% 10%
1
2
3
4
5
Coefficient Standard Error
Fact
or
26
27
0.00%
1.00%
2.00%
3.00%
0 1 2 3 4 5 6 7 8 9 10
Avg
Abs A
/E E
rror
Number of Factors
28
0%
10%
20%
30%
40%
50%
60%
70%
80%
90%
100%
0% 10%
20%
30%
40%
50%
60%
70%
80%
90%
100%
True
Pos
itive
Rat
e
False Positive Rate29
0
20,000
40,000
60,000
80,000
100,000
120,000
140,000
160,000
180,000
95%
96%
97%
98%
99%
100%
101%
102%
103%
104%
105%
1 2 3 4 5 6 7 8 Actu
als
A/E
Factor Xi
30
-40%
-30%
-20%
-10%
0%
10%
20%
30%
40%
0 10 20 30 40 50 60 70 80 90 100
Actu
al to
Mod
el A
vera
ge
Expected Deciles31
Cost-benefit of industry data
32
Example: variable annuity industry data
• 24 companies
• Seriatim monthly data for policyholder behavior and mortality
• January 2008 through December 2018
• $795 billion ending account value
33
How you measure value matters, and credibility is vastly improved with industry data
34
0%
25%
OTM 50+% OTM 25 -50%
OTM 5 -25%
ATM ITM 5 -25%
ITM 25 -50%
ITM 50 -100%
ITM 100%+
Surr
ende
r Rat
eGLWB
Shock -actuarial
Ultimate -actuarial
Shock -nominal
Ultimate -nominal
34
Industry data shows that GLWB income commencement is highest at issue and after bonuses expire…
35
0%
30%
1 2 3 4 5 6 7 8 9 10 11 12 13
Duration
Freq
uenc
y
35
0%
40%
ITM 100+% ITM 50 -100%
ITM 25 - 50% ITM 5 - 25% ATM OTM 5 - 25% OTM 25+%
Freq
uenc
y
…and that ultimate income commencement is dynamic
36
Dur 11+
Dur 3-10
36
0%
10%
<50 50-59 60-64 65-69 70-79 80-LEA Last Eligible
Annu
itiza
tion
Rate
Industry data shows that hybrid GMIB annuitization rates are backloaded…
3737
…and depend on economic value of other benefits, such as continued income utilization
38
0.0%
2.0%
<95% 95-100% 100-105% 105-110% 110-115% >115%
Annu
itiza
tion
Rate
Ratio of Income PV to Annuitization PV
38
Industry data also makes a better tabular mortality basis…
3939
0%
20%
40%
60%
80%
100%
120%
140%
0-49 50-54 55-59 60-64 65-69 70-74 75-79 80-84 85-89 90+
% o
f Tab
le
Actual vs. Ruark VAM 2015
Male Count Male Amount Female Count Female Amount Base
…and shows how income utilization affects mortality
40
0%
150%
First Year No Prior Withdrawals Prior LT and/or FullWDs only
Any Prior Excess WDs
% o
f Tab
le
Actual vs. RVAM 2015
Qualified
Non-qualified
Total
40
Modeling and assumptions
• Measuring goodness-of-fit for candidate models
• Testing predictive power on out-of-sample data
Using relevant industry data to improve candidate models
• Art + science: choosing, communicating, and ongoing recalibration
41
0
1,000
2,000
3,000
4,000
0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20
Baye
sian
Info
rmat
ion
Crite
rion
(BIC
)
Number of Factors
Company-only
Industry
42
0% 2% 4% 6% 8% 10%
1
2
3
4
5
6
7
8
9
10
Coefficient Standard Error
Fact
or
Company-only
Industry
43
44
0.00%
1.00%
2.00%
3.00%
0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20
Avg
Abs A
/E E
rror
Number of Factors
Industry
Company-only
Customize your model in a credibility-based framework
• Subject matter expertise
• Actuarial judgment
• Quantify the benefits of using relevant industry data
• Ongoing recalibration, so focus on the framework and its sense of range
45
46
0% 2% 4% 6% 8% 10%
1
2
3
4
5
6
7
8
9
10
Coefficient Standard Error
Fact
or
Industry
Customized blend
Company-only
47
0.00%
1.00%
2.00%
3.00%
0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20
Avg
Abs A
/E E
rror
Number of Factors
Company-only
Industry
Customized blend
How much is 1% A/E improvement worth to you?
Suppose 5.00% average annual surrender rates for your block
1% A/E improvement would be 0.05% annually and about 0.60% in present value terms
With 15% annualized market vol, hedge breakage (~2 s.d.) would be 0.18% of notionals
So what are your hedge notionals?
48
Hedge notionals Annualized hedge breakage (~ 2 s.d.)
$100 million $180,000
$1 billion $1,800,000
$10 billion $18,000,000
0.60% * 15% * 2
Cost-benefit of industry data
• Need to customize your model in a credibility-based framework
• Quantify the improvement in goodness-of-fit and predictive power metrics
• Quantify these improvements in financial terms – pricing margins, reserves, hedge breakage
• Quantify the cost to access and use relevant industry data
• Altogether, does this improve your financial risk profile?
Contrast this approach with unlocking ad nauseam
49
More data and/or
relevant industry data
Art + science, subject matter expertise and
actuarial judgment
More statistically justifiable
model factors and
dramatically improved fit
and predictive power
50
Discussion
51