Experience Rating Current Challenges

Presented by

Tony DiDonato, NCCI, Inc.

2003 CAS Seminar on Ratemaking

San Antonio, Texas

March 28, 2003

WCP-2

2

NCCI Perspective - Outline

I. ER Basics

II. Off-Balance in NCCI States

III. ERA – Experience Rating Adjustment

IV. Performance Testing Results for NCCI States

3

Actual / ExpectedAp + Ae(W) + Ee(1-W) +B

E + B

NCCI Plan - ER Basics

The formula:

A = Actual

E = Expected

p = primary

e = excess

B = Ballast

W = Weight

Some Qualifications:

• Premium thresholds

• Actual Losses limited

• Mods limited

• ERA

4

Countrywide* Off-Balance Factors

0.989

0.926 0.924 0.9230.931

0.937

0.954

0.941

0.964

0.924

0.90

0.91

0.92

0.93

0.94

0.95

0.96

0.97

0.98

0.99

1.00

1993 1994 1995 1996 1997 1998 1999 2000 2001 2002p

Year

Off-

Bal

ance

* NCCI states

5

Theoretical off-balance increases by about 0.0025 - 0.0035 for every percentage point of unanticipated frequency change.

Theoretical off-balance increases by about 0.002 - 0.003 for every percentage point of unanticipated severity change.

-20% -12% -4% 0% 4% 12% 20%

-20% 0.900 0.916 0.931 0.939 0.947 0.963 0.978

-12% 0.920 0.938 0.955 0.964 0.972 0.989 1.006

-4% 0.940 0.960 0.978 0.988 0.997 1.016 1.034

0% 0.951 0.970 0.990 1.000 1.010 1.029 1.049

4% 0.961 0.981 1.002 1.012 1.022 1.043 1.063

12% 0.981 1.003 1.025 1.036 1.047 1.069 1.091

20% 1.001 1.025 1.049 1.061 1.073 1.096 1.119

<-- Total Unanticipated % Change in Severity -->

Theoretical Off-balance

Tot

al U

nant

icip

ated

% C

hang

e in

Fre

quen

cy

Unanticipated Frequency and Severity: How is Off-Balance Impacted?

6

-20% -5% -1% 0% 1% 5% 20%

-20% 0.843 0.915 0.934 0.939 0.944 0.963 1.035

-5% 0.871 0.956 0.979 0.985 0.990 1.013 1.098

-1% 0.878 0.967 0.991 0.997 1.003 1.027 1.116

0% 0.880 0.970 0.994 1.000 1.006 1.030 1.120

1% 0.882 0.973 0.997 1.003 1.009 1.033 1.124

5% 0.890 0.984 1.009 1.015 1.022 1.047 1.141

20% 0.917 1.025 1.054 1.061 1.068 1.097 1.205

Theoretical off-balance increases by about 0.005 - 0.007 for every unanticipated additive percentage point in the D-ratio.

<-- Total Unanticipated Percentage Change in D-ratio -->

Theoretical Off-balance

Theoretical off-balance increases by about 0.002 - 0.004 for every percentage point of unanticipated losses.

To

tal U

na

ntic

ipa

ted

% C

ha

ng

e in

T

ota

l Lo

sse

s

Sensitivity of Off-Balance to:Total Losses vs. Primary/Excess Split

7

D-ratio Changes Over Time

• The primary change is due to severity trend

• Since severity generally increases over time, the d-ratio generally decreases

• ERA allows for a trended split point which would tend to stabilize the d-ratio, but trend has not yet been used on the split point

8

Primary Actual Primary Expected Difference BetweenExperience Total Actual / / Total Actual / Total Expected Actual D-ratio Average

Rating Years Total Expected (Actual D-ratio) (Expected D-ratio) Expected D-ratio Mod

1996 0.96

1997 0.95

1998 0.95

1999 0.94

2000 0.94

2001

0.84

0.85

0.87

0.84

0.88

0.89

0.30

0.30

0.30

0.27

0.26

0.25

0.28

0.29

0.30

0.28

0.28

0.28

0.02

0.02

0.00

(0.01)

(0.02)

(0.02) 0.95

Total Losses and Primary/Excess Split:Forecast vs. Actual Results

Intrastate Risks – NCCI States

9

• Countrywide (NCCI states) Off-Balance has increased in each of the last three years.

• Total actual losses that were less than total expected losses is the major reason for the relatively low Off-Balances from 1996-1999.

• While the split between claim frequency and severity did not drive the low Off-Balances, it did dampen the Off-Balance increases in the last three years.

• Unanticipated claim frequency has slightly more of an impact than does unanticipated severity on Mods.

Experience Rating Plan Off-Balance– Historical Analysis

10

• Manual loss ratios for the smallest premium sizes and for unrated risks are higher than the all-risk average.

• If the off-balance is 1.00, then there is no standard premium price differential between experience rated and unrated risks.

• Having an off-balance less than 1.00 can partially address the difference.

• The indicated standard premium level is still correct even if there is a net off-balance.

What Should the Experience Rating Plan Off-Balance Be ?

11

Developed Manual Loss Ratios byPolicy Manual Premium

(State X, Five Recent Policy Years)

0%

40%

80%

120%

160%

200%

$0 -

$50

0

$501

-$1

000

$1,0

01 -

$1,5

00

$1,5

01 -

$2,0

00

$2,0

01 -

$5,0

00

$5,0

01 -

$10,

000

$10,

001

-$2

5,00

0

$25,

001

-$5

0,00

0

$50,

001

-$1

00,0

00

$100

,001

-$1

50,0

00

$150

,001

-$2

00,0

00

$200

,001

-$3

00,0

00

$300

,001

-$5

00,0

00

$500

,001

and

Abo

ve

12

Developed Manual Loss Ratios (State X, Five Recent Policy Years)

0%

20%

40%

60%

80%

100%

120%

1995 1996 1997 1998 1999 1995-1999

Unrated Intrastate Interstate

13

Unrated Risks Rated Risks

% of Manual Premium

Manual Loss Ratio

Actual Loss / Expected Loss

Mod Needed to EqualizeStandard Premium Loss Ratios

Mod if Unrated Risksare Subsidized

10.0%

83.6%

1.439

1.000

1.000

90.0%

55.3%

0.951

0.661

0.951

Average

100.0%

58.1%

1.000

0.695

0.956

Simplified Hypothetical Illustration of Experience Rated vs. Unrated

Impact on Off-Balance

14

Summary of ERA Changes

(1)   Using only 30% of med-only claims in the Experience Rating formula

(2)   The Weighting Value (W) was increased

(3)   The primary/excess split point (currently $5,000) will be adjusted over time

ERA was designed to increase the incentive for employers to report small med-only claims and to improve the performance of the Plan. This was accomplished by the following three changes to the Plan: 

The effective date of ERA varies by state. The earliest effective date is 7/1/98, which is applicable in several states.

15

• Compared changes in the proportion of med-only claims in states adopting ERA vs. states that had not

• Changes in the average severity of med-only claims were also reviewed

• This review did not reveal a significant impact on the reporting of med-only claims due to ERA

ERA Impact on Med-Only Losses

16

% Chng in % Chng in% Chng in Med-Only % Chng in Med-Only

State Med-Only Severity State Med-Only SeverityA + 2.3% + 6.1% 1* + 0.1% + 14.6%B - 0.1% + 23.5% 2 + 1.1% + 21.0%C - 0.7% + 25.3% 3 + 8.5% + 15.1%D - 0.7% + 25.9% 4 + 1.2% + 24.0%E + 1.2% + 17.9% 5 + 3.7% + 9.9%F + 1.2% + 12.4% 6 + 3.4% + 21.9%G + 1.8% + 14.4% 7 - 4.4% - 19.5%H + 0.3% + 33.2% 8 + 1.3% + 18.9%I + 1.6% + 21.5% 9 + 1.3% + 26.3%

* State 1 reflects a 2-yr change

Med-Only Changes in Late 1990s

From policy period beginning 11/95 to policy period beginning 11/98

ERA States Non-ERA States

17

• 1998 mods were calculated for intrastate risks in the 9 states adopting ERA on 7-1-98

• Initially all rated under GERT, then all re-rated under ERA

• ERA showed slightly more accurate results

Performance Testing: GERT vs. ERA

18

Actual ActualSubsequent Subsequent

Quintile Losses LossesStratum Divided by Squared Deviation Divided by Squared Deviation

Determined Manual from Mean of (2) Modified from Mean of (4)by Mod Expected x 10,000 Expected x 10,000

(1) (2) (3) (4) (5)

1 0.70 900 0.93 121

GERT 2 0.85 225 1.00 16

3 0.94 36 1.04 04 1.09 81 1.07 95 1.48 2,304 1.12 64

Mean or Total 1.00 3,546 1.04 210

0.059

1 0.72 784 0.97 64

ERA 2 0.88 144 1.04 1

3 0.93 49 1.04 14 1.08 64 1.07 45 1.44 1,936 1.09 16

Mean or Total 1.00 2,977 1.05 86

0.029

Test Statistic: (5)/(3)=

Test Statistic: (5)/(3)=

Performance Testing: GERT vs. ERA

19

• Col (2): The actual losses are unlimited losses from WCSP data, generally at a 2nd report or subsequent.

• The rates/loss costs in effect during the appropriate time period were used as a proxy for expected losses. The column (2) ratios have been normalized to 1.00 to minimize differences between the actual and expected losses related to development, expenses, etc.

• Col (3) shows the deviation of each quintile group from the overall total.

• Col (4) reflects the normalization from Col (2), but after application of the Mod (in the denominator) the results were NOT re-normalized. This has no impact on the result in Col (5). The mean value shown is an intermediate step in the calculation and has no particular meaning.

• Col (5) shows the deviation of each quintile group from the overall total.

• The test statistics shown at the bottom of each analysis are key. A statistic less than 1.00 is expected from an Experience Rating Plan. Lower values of the statistic indicate better performance.

Brief Summary of Quintile Testing

20

GERT Individual State Quintiles

0.4

0.6

0.8

1.0

1.2

1.4

1.6

0.4 0.6 0.8 1.0 1.2 1.4 1.6 1.8

Actual / Expected Loss

Act

ua

l /

Mo

dif

ied

Exp

ecte

d L

oss

Performance Testing: GERT

Correlation = 0.654453

21

ERAIndividual State Quintiles

0.4

0.6

0.8

1.0

1.2

1.4

1.6

0.4 0.6 0.8 1.0 1.2 1.4 1.6 1.8

Actual / Expected Loss

Ac

tua

l /

Mo

dif

ied

Ex

pe

cte

d L

os

sPerformance Testing: ERA

Correlation = 0.543165

22

Countrywide* Quintile Testing Results

• Includes Interstate and Intrastate Risks

• Policy periods 7/1/97–6/30/98 and 7/1/98–6/30/99 were reviewed

• Two quintile groupings were tested:

– quintile groups which equalized risk count

– quintile groups which equalized expected losses

* NCCI states

23

Actual ActualSubsequent Subsequent

Quintile Losses LossesStratum Divided by Squared Deviation Divided by Squared Deviation Percentage of Percentage of

Determined Manual from Mean of (2) Modified from Mean of (4) Expected Loss Risks Countby Prior Mod Expected x 10,000 Expected x 10,000 in Quintile in Quintile

(1) (2) (3) (4) (5) (6) (7)

Expected Losses Uniformly Distributed Among Quintiles

1 0.66 1,129 1.04 26 20.0% 8.9%2 0.82 324 1.03 34 20.0% 20.0%3 0.95 28 1.07 3 20.0% 30.3%4 1.13 163 1.14 22 20.0% 18.3%5 1.43 1,880 1.13 14 20.0% 22.6%

Mean or Total 1.00 3,524 1.09 100 100.0% 100.0%Test Statistic: (5) / (3) = 0.028

Risk Count Uniformly Distributed Among Quintiles

1 0.72 772 1.04 29 33.4% 20.0%2 0.88 141 1.05 16 14.1% 20.0%3 0.99 0 1.10 1 13.8% 20.0%4 1.13 173 1.12 9 21.4% 20.0%5 1.47 2,234 1.13 17 17.3% 20.0%

Mean or Total 1.00 3,319 1.09 73 100.0% 100.0%Test Statistic: (5) / (3) = 0.022

Countrywide* Quintile Testing ResultsALL RISK SIZES

Policy Effective Period 7/1/97 - 6/30/98

* NCCI states

24

Countrywide* Quintile Testing ResultsALL RISK SIZES

Policy Effective Period 7/1/98 - 6/30/99

Actual ActualSubsequent Subsequent

Quintile Losses LossesStratum Divided by Squared Deviation Divided by Squared Deviation Percentage of Percentage of

Determined Manual from Mean of (2) Modified from Mean of (4) Expected Loss Risks Countby Prior Mod Expected x 10,000 Expected x 10,000 in Quintile in Quintile

(1) (2) (3) (4) (5) (6) (7)

Expected Losses Uniformly Distributed Among Quintiles

1 0.67 1,074 1.07 10 20.0% 8.9%2 0.82 329 1.04 30 20.0% 19.7%3 0.96 16 1.10 0 20.0% 30.4%4 1.12 138 1.12 7 20.0% 19.1%5 1.43 1,826 1.12 8 20.0% 22.0%

Mean or Total 1.00 3,382 1.10 55 100.0% 100.0%Test Statistic: (5) / (3) = 0.016

Risk Count Uniformly Distributed Among Quintiles

1 0.72 765 1.05 23 33.7% 20.0%2 0.89 112 1.07 6 14.9% 20.0%3 1.00 0 1.12 5 11.8% 20.0%4 1.13 168 1.12 6 21.9% 20.0%5 1.45 2,076 1.13 8 17.7% 20.0%

Mean or Total 1.00 3,121 1.10 48 100.0% 100.0%Test Statistic: (5) / (3) = 0.015

* NCCI states

25

0.4

0.6

0.8

1

1.2

1.4

1.6

0.25 0.5 0.75 1 1.25 1.5 1.75 2 2.25

Actual / Expected

Act

ual

/ M

od

ifie

d E

xpec

ted

(n

orm

aliz

ed b

y st

ate)

Individual State QuintilesALL RISK SIZES

Policy Effective Period 7/1/97 - 6/30/98

Correlation = 0.620352

26

0.4

0.6

0.8

1

1.2

1.4

1.6

0.25 0.5 0.75 1 1.25 1.5 1.75 2 2.25

Actual / Expected

Act

ual

/ M

od

ifie

d E

xpec

ted

(n

orm

aliz

ed b

y st

ate)

Correlation = 0.502986

Individual State QuintilesALL RISK SIZES

Policy Effective Period 7/1/98 - 6/30/99

27

Quintile* Test Quintile* TestStatistics for Statistics for

Expected Policy Period Policy PeriodLoss Range 7/1/97 - 6/30/98 7/1/98 - 6/30/99

$ 0 - $ 7,500 0.151 0.083

$ 7,500 - $ 15,000 0.051 0.060

$ 15,000 - $ 40,000 0.063 0.067

$ 40,000 - $ 100,000 0.040 0.027

$ 100,000 + 0.039 0.015

Quintiles by Risk CountTest Results by Size of Risk

* The risk count underlying each policy period and range is uniformly distributed among quintile stratum.

28

Quintiles by Expected LossesTest Results by Size of Risk

Quintile* Test Quintile* TestStatistics for Statistics for

Expected Policy Period Policy PeriodLoss Range 7/1/97 - 6/30/98 7/1/98 - 6/30/99

$ 0 - $ 7,500 0.155 0.080

$ 7,500 - $ 15,000 0.053 0.062

$ 15,000 - $ 40,000 0.064 0.066

$ 40,000 - $ 100,000 0.039 0.024

$ 100,000 + 0.040 0.009

* The expected loss volume underlying each policy period and range is uniformly distributed among quintile stratum.

29

Performance Testing: Next Steps

• Currently reviewing the 30% factor for Med-only claims under ERA

• Reviewing treatment of small risks

• Alternative performance measures are being reviewed

• Continued monitoring of ER Plan and performance