Generalized Linear Models II: Applying GLMs in Practice Duncan Anderson MA FIA Watson Wyatt LLP W W W. W A T S O N W Y A T T.
|
|
- Susanna Mitchell
- 6 years ago
- Views:
Transcription
1 Generalized Linear Models II: Applying GLMs in Practice Duncan Anderson MA FIA Watson Wyatt LLP W W W. W A T S O N W Y A T T. C O M
2 Agenda Introduction / recap Model forms and model validation Aliasing and "near aliasing" Interactions Smoothing, Combining models, Restrictions Tweedie GLMs Applications and interpreting the results
3 The premium rating process Rate level adjustments Expense loadings BI Freq x Amt = Cost 1 PD Freq x Amt = Cost 2 MED Freq x Amt = Cost 3 COL Freq x Amt = Cost 4 OTC Freq x Amt = Cost 5 Current Rates Competitor Model Risk Model Compare Lapse/takeup Model New Rates Model office
4 The premium rating process Rate level adjustments Expense loadings BI Freq x Amt = Cost 1 PD Freq x Amt = Cost 2 MED Freq x Amt = Cost 3 COL Freq x Amt = Cost 4 OTC Freq x Amt = Cost 5 Current Rates Competitor Model Risk Model Compare Lapse/takeup Model New Rates Model office
5 Generalized linear models E[Y i ] = i = g 1 ( X ij j + i ) Var[Y i ] =.V( i )/ i Consider all factors simultaneously Allow for nature of random process Robust and transparent Increasingly a global industry standard
6 Generalized linear models E[Y] = = g 1 (X. + ) Var[Y] =.V( )/
7 Generalized linear models E[Y] = = g 1 (X. + ) Link function Yvariate Design matrix Parameter estimates Offset term
8 Generalized linear models E[Y] = = g 1 (X. + ) Observed thing (data) Some function (user defined) Some matrix based on data (user defined) Parameters to be estimated (the answer!) Known effects
9 Generalized linear models Var[Y] =.V( )/ Scale parameter Variance function Prior weights Usually assume exponential family, eg = 2 (estimated), V(x) = 1 Var[Y i ] = 2 Normal = 1 (specifed), V(x) = x Var[Y i ] = i Poisson = k (estimated), V(x) = x 2 Var[Y i ] = k i2 Gamma
10 Link function g(x) Offset Error Structure V( ) Linear Predictor Form X. = i + j + k + l Data Y Scale Parameter Prior Weights Numerical MLE Parameter Estimates,,, Diagnostics
11 Maximum likelihood estimation Seek parameters which give highest likelihood function given data Likelihood Parameter 1 Parameter 2
12 Example of GLM output (real UK data).25 22% % % 7% 14 Log of multiplier % 5% 19% 4% 16% 2% 1 5% 19% 17% Exposure (policy years) Factor Expos ure On ew ay re la tiv ities Approx 2 SE from estimate Smoothed GL M estimate
13 "A Practitioner's Guide to Generalized Linear Models" CAS 24 Discussion Paper Programme Copies available here and at and
14 Data required Minimum 1, earned exposures Model separately by claim type Key fields for each risk period of earned exposure rating factors applicable at time (start of period) other "dummy" factors (area, time) incurred claim count (by type) incurred loss amount (by type), based on most recent estimates (optionally) earned premium on current basis
15 Preliminary analyses Oneway analyses Twoway analyses Correlation analyses Number of policies Distribution analyses High Vehicle Value Low New Old Vehicle Age Claim type 1 Third party property damage Vehicle type (Type) 25 2 Demonstration job Cla im type 2 Thir d par ty materi al damag e Wh ere Itpm^= an d Ntp m^= Level Number of records Exposure Premium Number of claims Incurred losses Claim frequency Average cost per claim Pure premium Loss ratio A 27,661 24,757 1,584,626 1,87 8,457,28 7.3% 4, % B 22,89 19,777 9,623,698 1,598 6,957, % 4, % C 13,768 12,334 6,35,96 1,11 4,245,92 8.2% 4, % D 19,662 17,592 9,382,767 1,584 6,7,943 9.% 3, % E 11,235 1,76 5,676, ,262, % 3, % F 5,67 5,37 3,118, ,858, % 3, % Numbe r o f claims ,22 89,572 44,691,424 7,532 3,852, % 4, % Average claim size
16 Agenda Introduction / recap Model forms and model validation Aliasing and "near aliasing" Interactions Smoothing, Combining models, Restrictions Tweedie GLMs Applications and interpreting the results
17 Interesting properties Poisson multiplicative parameter estimates unchanged if group by unique combination of rating factor invariant to measures of time Gamma multiplicative parameter estimates unchanged by grouping but standard errors are not generally do not group except for multiple claims on a risk in a policy period invariant to measures of currency Logistic (binomial / logit) maps (,1) to (, ) invariant to whether measuring of success / failure (eg same if model lapse / renew) more appropriate for retention/conversion analyses, but harder to communicate
18 Typical model forms Y Claim frequency Claim number Average claim amount Probability (eg lapses) g(x) ln(x) ln(x) ln(x) ln(x/(1x)) Error Poisson Poisson Gamma Binomial V(x) 1 x 1 x estimate x 2 1 x(1x) exposure 1 # claims 1 ln(exposure)
19 Example of effect of changing assumed error Data Normal
20 Example of effect of changing assumed error Data Normal Poisson
21 Example of effect of changing assumed error Data Normal Poisson Gamma
22 Example of effect of changing assumed error 2 Example portfolio with five rating factors, each with five levels A, B, C, D, E Typical correlations between those rating factors Assumed true effect of factors Claims randomly generated (with Gamma) Random experience analysed by three models
23 Example of effect of changing assumed error Log of multiplier A B C D E.2 True effect
24 Example of effect of changing assumed error Log of multiplier A B C D E.2 True effect One way
25 Example of effect of changing assumed error Log of multiplier A B C D E.2 True effect One way GLM / Normal
26 Example of effect of changing assumed error Log of multiplier A B C D E.2 True effect One way GLM / Normal GLM / Gamma
27 Example of effect of changing assumed error Log of multiplier A B C D E.2 True effect GLM / Gamma + 2 SE 2 SE
28 Model testing Use only those factors which are predictive standard errors of parameter estimates F tests / 2 tests on deviances stepwise approach (helpful if used with care) consistency over time human intuition Make sure the model is reasonable residual plots (histograms / QQ / residual vs fitted value etc) leverage / Cook's distance BoxCox
29 Standard errors Roughly speaking, for a parameter p: SE = 1 / ( 2 / p 2 Likelihood) Likelihood Parameter 1 Parameter 2
30 GLM output (significant factor) % 138% % 14 Log of multiplier % 39% 45% 58% 84% 73% 72% 93% Exposure (years).2 6 % 5% Vehicle symbol Onew ay relativities Approx 95% confidence interval Parameter es timate P value =.%
31 GLM output (insignificant factor) % 14 Log of multiplier.5.5 % 4% 1% 3% 5% % 5% 1% 3% 1% 4% 1% Exposure (years) Vehicle symbol Onew ay relativities Approx 95% confidence interval Par ameter estimate P value = 52.5%
32 Awkward cases Log of multiplier % 5% 16% 14% 11% 1% 22% % 42% 65% 92% 11% 123% 159% 2%
33 Awkward cases Log of multiplier % 35% 28% 22% 16% 11% 5% % 5% 1% 14% 18% 22% 26% 3%
34 Deviance type III tests Single figure measure of goodness of fit Try model with & without a factor Statistical tests show the theoretical significance given the extra parameters Age Sex Vehicle Zone Model A Fitted value Deviance = 9585 df = Excess NCD? Age Sex Vehicle Model B Fitted value Deviance = 964 df = Excess NCD
35 GLM output (significant factor) % 138% % 14 Log of multiplier % 39% 45% 58% 84% 73% 72% 93% Exposure (years).2 6 % 5% Vehicle symbol Onew ay relativities Approx 95% confidence interval Parameter es timate P value =.%
36 GLM output (insignificant factor) % 14 Log of multiplier.5.5 % 4% 1% 3% 5% % 5% 1% 3% 1% 4% 1% Exposure (years) Vehicle symbol Onew ay relativities Approx 95% confidence interval Par ameter estimate P value = 52.5%
37 Consistency over time A B C D A B C D
38 Consistency over time % 1 Log of multiplier % 27% 45% 38% 34% 51% 44% 4% 63% 56% 97% 81% 73% 78% 76% 64% 81% 69% 65% 14% 89% 86% 18% 14% 15% 15% 149% 118% 14% Exposure (years).2 4% % 4% 1% 5% 2% Vehicle symbol.year of exposure Approx 95% confidence interval, Year of exposure: 2 Approx 95% confidence interval, Year of exposure: 21 Approx 95% confidence interval, Year of exposure: 22 Parameter estimate, Year of exposure: 2 Parameter estimate, Year of exposure: 21 Parameter estimate, Year of exposure: 22
39 Consistency over time % % 8 Log of multiplier % 21% 23% 2% 13% 2% 29% 15% 25% 3% 4% % 26% 14% 9% 16% 15% 3% 22% Exposure (years) Territory.Year of exposure Approx 95% confidence interval, Year of exposure: 2 Approx 95% confidence interval, Year of exposure: 21 Approx 95% confidence interval, Year of exposure: 22 Smoothed estimate, Year of exposure: 2 Smoothed estimate, Year of exposure: 21 Smoothed estimate, Year of exposure: 22
40 Intuition Are ordered categorical variables well behaved? Can you believe it, given correlations with other factors? Can the underwriters believe it? How different is it to the oneway? What does this factor do in other frequency/amounts models and for other claim types?
41 Practical model iteration Start with all factors if possible In theory reject one at a time In practice be more brutal if oneways are similar to GLM parameter estimates Recheck excluded factors at end Interactions Stepwise algorithms can be useful if used with care This is what takes time
42 Model testing Use only those factors which are predictive standard errors of parameter estimates F tests / 2 tests on deviances stepwise approach (helpful if used with care) consistency over time human intuition Make sure the model is reasonable residual plots (histograms / QQ / residual vs fitted value etc) leverage / Cook's distance BoxCox
43 Residuals Residuals Fitted values Data
44 Residuals Several forms, eg standardized deviance sign (Y u u ) / ( (1h u ) ) ½ 2 u ( Y u ) / V( d standardized Pearson Y u u (.V( ).(1h ) / ) u u u Standardized deviance Normal (,1) Numbers/frequency residuals problematical ½ Y u u
45 Residuals Histogram of Deviance Residuals Run 12 (Final models with analysis) Model 8 (AD amounts) Frequency Size of deviance residuals Pretium 8/1/24 12:24
46 Residuals Residual Pretium 4/5/24 11:3 Log of fitted value
47 Residuals
48 Gamma data, Gamma error Plot of deviance residual against fitted value Run 12 (All claim types, final models, N&A) Model 6 (Own damage, Amounts) 2 1 Deviance Residual Pretium 8/1/24 12:32 Fitted Value
49 Gamma data, Normal error Plot of deviance residual against fitted value Run 12 (All claim types, final models, N&A) Model 7 (Own damage, Amounts) Deviance Residual Pretium 8/1/24 12:32 Fitted Value
50 Leverage Plot of leverage against fitted value Run 12 (All claim types, final models, N&A) Model 6 (Own damage, Amounts) Leverage Pretium 8/1/24 12:32 Fitted Value
51 BoxCox link function investigation GLM structure is E[Y] = = g 1 (X. + ) Var[Y] =.V( ) / Box Cox transforms defines g(x) = ( x 1 ) / for, ln(x) for = = 1 g(x) = x 1 additive (with base level shift) g(x) ln(x) multiplicative (via maths) = 1 g(x) = 1 1/x inverse (with base level shift) Try different values of and measure goodness of fit to see which fits experience best
52 BoxCox link function investigation Motor third party property frequencies Likelihood Inverse Multiplicative Additive
53 BoxCox link function investigation Motor third party property average amounts Likelihood Inverse Multiplicative Additive
54 BoxCox link function investigation Comparing fitted values of different link functions 35 Not much 3 25 Count of records Ratio of fitted values for Lambda = (mult) to Lambda =.3
55 Agenda Introduction / recap Model forms and model validation Aliasing and "near aliasing" Interactions Smoothing, Combining models, Restrictions Tweedie GLMs Applications and interpreting the results
56 Aliasing and "near aliasing" Aliasing the removal of unwanted redundant parameters Intrinsic aliasing occurs by the design of the model Extrinsic aliasing occurs "accidentally" as a result of the data
57 Example Suppose we wanted a model of the form: = + if age < if age if age > if sex male 1 + if sex female 2
58 Form of X. in this case Age Sex <3 34 >4 M F
59 Example Suppose we wanted a model of the form: = + if age < if age "Base levels" + if age > if sex male 1 + if sex female 2
60 X. having adjusted for base levels Age Sex <3 34 >4 M F
61 Intrinsic aliasing Example job Run 16 Model 3 Small interaction Third party material damage, Numbers % 138%.8 2 Log of multiplier % 28% 24% 15 1 Exposure (years).2 % 19% 6% 11% Age of driver Onew ay relativities Approx 95% confidence interval Unsmoothed estimate Smoothed estimate
62 Extrinsic aliasing If a perfect correlation exists, one factor can alias levels of another Eg if doors declared first: Selected base Exposure: # Doors Unknown Colour Selected base Red 13,234 12,343 13,432 13,432 Green 4,543 4,543 13,243 2,345 Blue 6,544 5,443 15,654 4,565 Black 4,643 1,235 14,565 4,545 Further aliasing Unknown 3,242 This is the only reason the order of declaration can matter (fitted values are unaffected)
63 Extrinsic aliasing Example job Run 16 Model 3 Small interaction Third party material damage, Numbers 1 135% 6 Log of multiplier % 39% 45% 57% 82% 72% 7% 91% 13% Exposure (years) % % Malus No claim discount Onew ay relativities Approx 95% confidence interval Unsmoothed estimate Smoothed estimate
64 "Near aliasing" If two factors are almost perfectly, but not quite aliased, convergence problems can result and/or results can become hard to interpret Selected base Exposure: # Doors Unknown Colour Selected base Red 13,234 12,343 13,432 13,432 Green 4,543 4,543 13,243 2,345 Blue 6,544 5,443 15,654 4,565 Black 4,643 1,235 14,565 4,545 2 Unknown 3,242 Eg if the 2 black, unknown doors policies had no claims, GLM would try to estimate a very large negative number for unknown doors, and a very large positive number for unknown colour
65 Agenda Introduction / recap Model forms and model validation Aliasing and "near aliasing" Interactions Smoothing, Combining models, Restrictions Tweedie GLMs Applications and interpreting the results
66 No interaction age Example job Run 12 Model 3 Small interaction Third party material damage, Numbers %.8 15% % Log of multiplier % 24% % 6% 11% 15 1 Exposure.2 2% Age of driver Approx 2 SEs from estimate Unsmoothed estimate Smoothed estimate P level =.% Rank 7/7
67 No interaction sex Example job Run 12 Model 3 Small interaction Third party material damage, Numbers.3 % Log of multiplier Exposure % Female Male Sex of driver Approx 2 SEs from estimate Unsmoothed estimate Smoothed estimate P level =.% Rank 2/7
68 Age sex interaction Example job Run 5 Model 3 Small interaction Third party material damage, Numbers 1 155% 138% % 63% Log of multiplier % 4% 28% 19% 24% 2% 2% % 6% 6% 13% 11% Exposure.2 18% 19% Age of driver.sex of driver P level =.% Rank 6/6 Approx 2 SEs from estimate, Sex of driver: Female Approx 2 SEs from estimate, Sex of driver: Male Unsmoothed estimate, Sex of driver: Female Unsmoothed estimate, Sex of driver: Male Smoothed estimate, Sex of driver: Female Smoothed estimate, Sex of driver: Male
69 Interactions W x X Y x Z x No interaction A B C D x x x Full interaction A B C D W x x x x X x x x Y x x x x Z x x x x Marginal interaction A B C D x x x W x x x x X Y x x x x Z x x x x
70 Interactions Factor 1: A B C D Full interaction Factor 2: W X Y Z Factor 1: A B C D Marginal interaction Factor 2: W X Y Z Consider D/Z: 2.1 = 1.2 * 1.4 * 1.25
71 Age sex interaction Example job Run 5 Model 3 Small interaction Third party material damage, Numbers 1 155% 138% % 63% Log of multiplier % 4% 28% 19% 24% 2% 2% % 6% 6% 13% 11% Exposure.2 18% 19% Age of driver.sex of driver P level =.% Rank 6/6 Approx 2 SEs from estimate, Sex of driver: Female Approx 2 SEs from estimate, Sex of driver: Male Unsmoothed estimate, Sex of driver: Female Unsmoothed estimate, Sex of driver: Male Smoothed estimate, Sex of driver: Female Smoothed estimate, Sex of driver: Male
72 Marginal interaction: Age effect Example job Run 16 Model 3 Small interaction Third party material damage, Numbers % 138%.8 2 Log of multiplier x 63% 28% 24% 15 1 Exposure (years).2 % 19% 6% 11% Age of driver Onew ay relativities Approx 95% confidence interval Unsmoothed estimate Smoothed estimate
73 Marginal interaction: Age.Sex (ie additional female multipliers) Example job Run 16 Model 3 Small interaction Third party material damage, Numbers Log of multiplier x.64 = 1.63 as before % % % % % 3% 7% 14% % 2% 1% % 13% % 27% % Exposure (years) % 38% Age of driver.sex of driver Approx 95% confidence interval, Sex of driver: Female Unsmoothed estimate, Sex of driver: Female Unsmoothed estimate, Sex of driver: Male Smoothed estimate, Sex of driver: Female Smoothed estimate, Sex of driver: Male
74 An example of no interaction.8 1% 45 91% 82% % 56% 6% 46% 66% 51% 46% 39% 5% 35% 51% 35 3 Log of multiplier.2 22% 2% 17% 7% % % 5% 25 2 Exposure (years) 6% 7% 7% 6% % 31% Complete interaction of Age of driver and Payment frequency P value =.% Approx 95% confidence interval, Payment frequency: Yearly Approx 95% confidence interval, Payment frequency: Halfyearly Approx 95% confidence interval, Payment frequency: Quarterly Parameter estimate, Payment frequency: Yearly Parameter estimate, Payment frequency: Halfyearly Parameter estimate, Payment frequency: Quarterly
75 An example of no interaction % 46% 51% 2 Log of multiplier.2 22% 2% % 7% 7% 15 1 Exposure (years).2 22% Age of driver Oneway relativities Approx 95% confidence interval Parameter estimate P value =.%
76 An example of no interaction % Log of multiplier % 3 2 Exposure (years).5 % 1.5 Yearly Halfyearly Quarterly Payment frequency Oneway relativities Approx 95% confidence interval Parameter estimate P value =.%
77 An example of no interaction % 35 Log of multiplier % 2% % 6% 2% % 8% 3% % 7% 2% % 5% 6% % % 12% 9% % 5% % % 6% Exposure (years) % Marginal interaction of Age of driver and Payment frequency P value = 61.9% Approx 95% confidence interval, Payment frequency: Halfyearly Approx 95% confidence interval, Payment frequency: Quarterly Parameter estimate, Payment frequency: Yearly Parameter estimate, Payment frequency: Halfyearly Parameter estimate, Payment frequency: Quarterly
78 Why marginal interactions can be hard to interpret Exposure ('s) A B C D A 98 1,31 1,42 87 B 1,3 1,84 1,34 1 C 2,62 3,58 2,11 2,43 D 4,8 4,76 2,3 1,68 With low frequency Marginal interaction parameter estimates A B C D A B C D
79 Interactions Age Factor Group Factor
80 Interactions Group > Age v
81 Interactions Group Factor Age Factor
82 Interactions testing consistency with time Testing consistency with time Run 1 Model 1 Interaction with time Third party property damage (numbers) Log of multiplier Exposure (years) < Policyholder age.calendar year P value = 72.2% Rank 3/1 Approx 95% confidence interval, Calendar year: 21 Approx 95% confidence interval, Calendar year: 22 Approx 95% confidence interval, Calendar year: 23
83 Agenda Introduction / recap Model forms and model validation Aliasing and "near aliasing" Interactions Smoothing, Combining models, Restrictions Tweedie GLMs Applications and interpreting the results
84 Smoothing Models can often 15 be improved by 1 smoothing raw 5 statistical parameter 5 estimates where factors have a natural order Log of multiplier Best done at frequency / severity level Artificial constraints and commercial smoothing come later
85 Combining claim elements I BI Freq x Amt = Cost 1 PD Freq x Amt = Cost 2 MED Freq x Amt = Cost 3 Multiply factors for frequencies and amounts Calculate risk premium as sum of claim elements COL Freq x Amt = Cost 4 OTC Freq x Amt = Cost 5
86 Combining claim elements II Consider current exposure BI Freq x Amt = Cost 1 PD Freq x Amt = Cost 2 MED Freq x Amt = Cost 3 COL Freq x Amt = Cost 4 OTC Freq x Amt = Cost 5 Calculate expected frequency and amount for each claim type for each record Combine to give expected total cost of claims for each record Fit model to this expected value
87 Calculation of risk premium TPPD Numbers TPPD Amounts TPBI Numbers TPBI Amounts Intercept 32% 1 12% 486 Sex Male Female Area Town Country Policy Sex Area WWNUM1 WWAMT1 WWNUM2 WWAMT2 WWCC1 WWCC2 WWRSKPRM M T 32% 1 12% F T 24% 12 8% M C 4% 7 9% F C 3% 84 6%
88 Risk premium standard errors Risk premium model standard errors are small owing to the smoothness of the expected value It is possible to approximate standard error of risk premium parameter estimates based on standard errors of parameter estimates in underlying models Care needed in interpretting such approximations since they do not reflect model error, eg deciding to exclude a marginal factor
89 Restricted models Log of multiplier Restriction Log of multiplier BonusMalus Vehicle group Log of multiplier.5 1 Log of multiplier A B C D E F G H I J K L M N O Geographical zone Age of driver
90 Restricted models Log of multiplier Restriction Log of multiplier BonusMalus Vehicle group Log of multiplier Log of multiplier 1.5 Model compensates (as best it can) to allow for restriction A B C D E F G H I J K L M N O Geographical zone Age of driver
91 Restricted models E[Y] = = g 1( X + ) Offset Offset term used for known effects, eg exposure in a numbers model Can also be used to constrain model (eg claim free years / payment frequency / amount of cover) Other factors adjusted to compensate
92 Testing the effectiveness of restrictions 12 1 Effect of restricting Bonus Malus Other factors not very correlated with Bonus Malus 8 Count of records Ratio A B C D E F G H I J K L M N O P Q R S
93 Testing the effectiveness of restrictions 12 1 Effect of restricting Bonus Malus Adding in a measure of claim free years 8 Count of records Ratio A B C D E F G H I J K L M N O P Q R S
94 Agenda Introduction / recap Model forms and model validation Aliasing and "near aliasing" Interactions Smoothing, Combining models, Restrictions Tweedie GLMs Applications and interpreting the results
95 .2 Tweedie distributions Incurred losses have a point mass at zero and then a continuous distribution Poisson and gamma not suited to this Tweedie distribution has point mass and parameters which can alter the shape to be like Poisson and gamma above zero f Y ( y;,, ) n 1 1 ( ) ( 1/ y) ( n ) n! y p( Y ) exp ( ) n.exp [ y ( )] for y
96 Tweedie distributions Tweedie = k, V(x) = x p Var[Y] = k p p=1 corresponds to Poisson, p=2 to gamma Defines a valid distribution for p<, 1<p<2, p>2 Can be considered as Poisson/gamma process for 1<p<2 Need to estimate both k and p when fitting models often estimate a where p = (2a)/(1a) Typical values of p for insurance incurred claims around, or just under, 1.5
97 Example 1: frequency Comparison of Tweedie model with traditional frequency/amounts approach Run 7 Model 2 Frequency % 446% 1 Log of multiplier % 128% 94% 96% 44% % 23% 15% 21% 17% Exposure (years) % A B C D E F G H I J K L M Bonus Malus Onew ay relativities Approx 95% confidence interval Unsmoothed estimate Smoothed estimate P v alue =.% Rank 12/12
98 Example 1: amounts Comparison of Tweedie model with traditional frequency/amounts approach Run 7 Model 6 Amounts Log of multiplier % 26% 6% 1% 5% 3% 7% % 2% 3% 2% 4% 1% Number of claims A B C D E F G H I J K L M Bonus Malus EXCLUDED FACTOR Onew ay relativities Approx 95% confidence interval Unsmoothed estimate Smoothed estimate P v alue = 5.9% Rank 4/12
99 Example 1: traditional RP vs Tweedie Comparison of Tweedie model with traditional frequency/amounts approach Run 11 Model 2 Tweedie Models % 568% 443% 446% 1 Log of multiplier % 128% 94% 96% 44% % 23% 15% 21% 17% Exposure (years) % A B C D E F G H I J K L M Bonus Malus Tw eedie SE Tw eedie RPSE RP
100 Example 2: traditional RP vs Tweedie Comparison of Tweedie model with traditional frequency/amounts approach Run 11 Model 1 Tweedie Models.1 45 Log of multiplier % 9% % Exposure (years) Category 1 Category 2 Occupation Tw eedie SE Tw eedie RPSE RP Numbers Amounts
101 Example 3: frequency Comparison of Tweedie model with traditional frequency/amounts approach Run 7 Model 1 Frequency.4 8 Log of multiplier % 6% 2% % 2% 2% 5% 8% 3% 8% 8% 3% 17% 2% Exposure (years) < Unknown 39% 1 Age of driver Onew ay relativities Approx 95% confidence interval Unsmoothed estimate Smoothed estimate P value =.% Rank 5/12
102 Example 3: amounts Comparison of Tweedie model with traditional frequency/amounts approach Run 7 Model 5 Amounts.18 16% 6 Log of multiplier % 3% 2% % 1% 1% 6% 9% 5% 3% 5% % 2% 4% Number of claims < Unknown Age of driver EXCLUDED FACTOR Onew ay relativities Approx 95% confidence interval Unsmoothed estimate Smoothed estimate P value = 5.6% Rank 4/9
103 Example 3: traditional RP vs Tweedie Comparison of Tweedie model with traditional frequency/amounts approach Run 11 Model 1 Tweedie Models Log of multiplier % 23% 11% 6% 2% 1% % 2% 4% 2% % 13% 5% 2% 8% 3% 1% 16% 8% 15% 8% 2% 3% 14% 17% 16% 2% 28% Exposure (years).4 39% 1.6 < Unknown Age of driver Tw eedie SE Tw eedie RPSE RP Numbers
104 Agenda Introduction / recap Model forms and model validation Aliasing and "near aliasing" Interactions Smoothing, Combining models, Restrictions Tweedie GLMs Applications and interpreting the results
105 Comparison with actual rates Demonstration job Run 1 Model 2 Third party material, standard risk premium run Unsmoothed standard risk premium model.8 89% % 67% 2.4 Log of multiplier.2 22% 22% 11% 3% % 22% % % % % % 1% 15 1 Exposure.2 22% 1% 18% MAGE Age of driver Approx 2 SEs from unsmoothed estimate Unsmoothed unr estricted estimate Unsmoothed restricted estimate Current rating structure
106 Comparison with actual rates Demonstration job Run 1 Model 2 Third party material, standard risk premium run Unsmoothed standard risk premium model % Log of multiplier % 15% 18% 7% 1% % % 22% 8% 49% 11% 31% Exposure % to to 17 MGROUP Group of vehicle Approx 2 SEs from unsmoothed estimate Unsmoothed unr estricted estimate Unsmoothed restricted estimate Current rating structure
107 Comparison with actual rates Demonstration job Run 1 Model 2 Third party material, standard risk premium run Unsmoothed standard risk premium model % 49% 2 Log of multiplier % 5% 4% 2% % % 4% 5% 1% 7% 13% 11% 15% 16% Exposure.2 A B C D E F G H MAREA Area of garage Approx 2 SEs from unsmoothed estimate Unsmoothed unrestricted estimate Unsmoothed restricted estimate Curr ent r ating str ucture
108 Comparison with actual rates Demonstration job Run 1 Model 2 Third party material, standard risk premium run Unsmoothed standard risk premium model % 5 4 Log of multiplier % 3 2 Exposure.5 % % 5% 5% 1.5 Yearly Halfyearly Quarterly MPFREQ Payment frequency Approx 2 SEs from unsmoothed estimate Unsmoothed unrestricted estimate Unsmoothed restricted estimate Current rating structure
109 Impact analysis 7 Currently profitable business Example job 6 5 Count of records Currently unprofitable business Ratio: Risk Premium / Current tariff
110 Impact analysis Example job 7 18% 17% Count of records % 15% 14% 13% 12% 11% 1% 9% 8% 7% 6% 5% 4% Loss ratio % Ratio: Risk Premium / Current tariff Yearly Claims / Earnedprem
111 Impact analysis Example job Age of driver 7 18% 17% Count of records % 15% 14% 13% 12% 11% 1% 9% 8% 7% 6% 5% 4% Loss ratio % Ratio: Risk Premium / Current tariff Claims / Earnedprem
112 Impact analysis Example job Area of garage 7 18% 17% Count of records % 15% 14% 13% 12% 11% 1% 9% 8% 7% 6% 5% 4% Loss ratio % Ratio: Risk Premium / Current tariff A B C D E F G H Claims / Earnedprem
113 Impact analysis Example job Class of vehicle 7 18% 17% Count of records % 15% 14% 13% 12% 11% 1% 9% 8% 7% 6% 5% 4% Loss ratio % Ratio: Risk Premium / Current tariff A & Y & Z B C D & E F & G & H & J Claims / Earnedprem
114 Impact analysis Example job Payment frequency 7 18% 17% 6 16% 15% 5 14% 13% Count of records Good proof of effectiveness of model 12% 11% 1% 9% 8% 7% 6% 5% Loss ratio 4% % Ratio: Risk Premium / Current tariff Yearly Halfyearly Quaterly Claims / Earnedprem
115 The premium rating process Rate level adjustments Expense loadings BI Freq x Amt = Cost 1 PD Freq x Amt = Cost 2 MED Freq x Amt = Cost 3 COL Freq x Amt = Cost 4 OTC Freq x Amt = Cost 5 Current Rates Competitor Model Risk Model Compare Lapse/takeup Model New Rates Model office
116 Considering the competitive position Example of competitor analysis Third party cover Log of multiplier % 5% 45% 31% 43% 4% 28% 4% 35% 24% 28% 3% 2% 23% 25% 18% 19% 2% 1% 17% 15% 3% 4% 1% 5% 9% 5% % % 11% 7% 5% 28% 15% 1% 35% 18% 15% 48% 31% 2% 62% 3% 25% 7% 48% 3% 8% 41% 35% 91% 44% 4% 17% 65% 45% 125% 82% 5% Exposure (years) Vehicle group Current tariff Approx 95% confidence interval Third cheapest market quote Smoothed estimate P value =.% Rank 9/11
117 Profitability (theoretical premium/current premium) Profitable Considering the competitive position Competitiveness (market premium/current premium) Competitive
118 Modeling retention / conversion Age Sex Vehicle age Premium Claims Premium / Competitors' premium GLM Probability Model normal factors other products held payment method change in cover discount expectation plus source change in premium claims history competitiveness
119 The premium rating process Rate level adjustments Expense loadings BI Freq x Amt = Cost 1 PD Freq x Amt = Cost 2 MED Freq x Amt = Cost 3 COL Freq x Amt = Cost 4 OTC Freq x Amt = Cost 5 Current Rates Competitor Model Risk Model Compare Lapse/takeup Model New Rates Model office
120 Profitability scoring Construct profitability score based on expected loss ratio Profitability score can then be used to target sections of a portfolio Expected loss ratio can be modeled using a risk premium model offset by current premium rates Expected loss ratio can be banded into discrete bands if desired
121 Profitability scoring Distribution of score 25 16% 14% 2 12% Number of policies % 8% 6% Actual loss ratio 5 4% 2% Score based on expected loss ratio % Number of policies Actual loss ratio
122 Generalized Linear Models II: Applying GLMs in Practice Duncan Anderson MA FIA Watson Wyatt LLP W W W. W A T S O N W Y A T T. C O M
Agenda. Current method disadvantages GLM background and advantages Study case analysis Applications. Actuaries Club of the Southwest
watsonwyatt.com Actuaries Club of the Southwest Generalized Linear Modeling for Life Insurers Jean-Felix Huet, FSA November 2, 29 Agenda Current method disadvantages GLM background and advantages Study
More informationGLM III - The Matrix Reloaded
GLM III - The Matrix Reloaded Duncan Anderson, Serhat Guven 12 March 2013 2012 Towers Watson. All rights reserved. Agenda "Quadrant Saddles" The Tweedie Distribution "Emergent Interactions" Dispersion
More informationPredictive Modeling GLM and Price Elasticity Model. David Dou October 8 th, 2014
Predictive Modeling GLM and Price Elasticity Model David Dou October 8 th, 2014 History of Predictive Modeling Pre-Computer Era: Triangles on a giant spreadsheet PC Era: Microsoft Excel oneway relativities
More informationThe Matrix Inverted A Primer in GLM Theory and Practical Issues. March 11-12, 2004 CAS Ratemaking Seminar Roosevelt Mosley, FCAS, MAAA
The Matrix Inverted A Primer in GLM Theory and Practical Issues March 1112, 2004 CAS Ratemaking Seminar Roosevelt Mosley, FCAS, MAAA Practical Issues Data Analysis Implementation Data Data Topics How much?
More informationQuantile Regression. By Luyang Fu, Ph. D., FCAS, State Auto Insurance Company Cheng-sheng Peter Wu, FCAS, ASA, MAAA, Deloitte Consulting
Quantile Regression By Luyang Fu, Ph. D., FCAS, State Auto Insurance Company Cheng-sheng Peter Wu, FCAS, ASA, MAAA, Deloitte Consulting Agenda Overview of Predictive Modeling for P&C Applications Quantile
More informationSession 178 TS, Stats for Health Actuaries. Moderator: Ian G. Duncan, FSA, FCA, FCIA, FIA, MAAA. Presenter: Joan C. Barrett, FSA, MAAA
Session 178 TS, Stats for Health Actuaries Moderator: Ian G. Duncan, FSA, FCA, FCIA, FIA, MAAA Presenter: Joan C. Barrett, FSA, MAAA Session 178 Statistics for Health Actuaries October 14, 2015 Presented
More informationAnd The Winner Is? How to Pick a Better Model
And The Winner Is? How to Pick a Better Model Part 2 Goodness-of-Fit and Internal Stability Dan Tevet, FCAS, MAAA Goodness-of-Fit Trying to answer question: How well does our model fit the data? Can be
More informationINSTITUTE AND FACULTY OF ACTUARIES. Curriculum 2019 SPECIMEN EXAMINATION
INSTITUTE AND FACULTY OF ACTUARIES Curriculum 2019 SPECIMEN EXAMINATION Subject CS1A Actuarial Statistics Time allowed: Three hours and fifteen minutes INSTRUCTIONS TO THE CANDIDATE 1. Enter all the candidate
More informationLoss Cost Modeling vs. Frequency and Severity Modeling
Loss Cost Modeling vs. Frequency and Severity Modeling 2013 CAS Ratemaking and Product Management Seminar March 13, 2013 Huntington Beach, CA Jun Yan Deloitte Consulting LLP Antitrust Notice The Casualty
More informationSession 5. A brief introduction to Predictive Modeling
SOA Predictive Analytics Seminar Malaysia 27 Aug. 2018 Kuala Lumpur, Malaysia Session 5 A brief introduction to Predictive Modeling Lichen Bao, Ph.D A Brief Introduction to Predictive Modeling LICHEN BAO
More informationWC-5 Just How Credible Is That Employer? Exploring GLMs and Multilevel Modeling for NCCI s Excess Loss Factor Methodology
Antitrust Notice The Casualty Actuarial Society is committed to adhering strictly to the letter and spirit of the antitrust laws. Seminars conducted under the auspices of the CAS are designed solely to
More informationUse of GLMs in a competitive market. Ji Yao and Simon Yeung, Advanced Pricing Techniques (APT) GIRO Working Party
Use of GLMs in a competitive market Ji Yao and Simon Yeung, Advanced Pricing Techniques (APT) GIRO Working Party 12 June 2013 About the presenters Dr. Ji Yao is a manager with Ernst & Young s EMEIA insurance
More informationInstitute of Actuaries of India Subject CT6 Statistical Methods
Institute of Actuaries of India Subject CT6 Statistical Methods For 2014 Examinations Aim The aim of the Statistical Methods subject is to provide a further grounding in mathematical and statistical techniques
More informationSubject CS1 Actuarial Statistics 1 Core Principles. Syllabus. for the 2019 exams. 1 June 2018
` Subject CS1 Actuarial Statistics 1 Core Principles Syllabus for the 2019 exams 1 June 2018 Copyright in this Core Reading is the property of the Institute and Faculty of Actuaries who are the sole distributors.
More informationLogit Models for Binary Data
Chapter 3 Logit Models for Binary Data We now turn our attention to regression models for dichotomous data, including logistic regression and probit analysis These models are appropriate when the response
More informationSession 5. Predictive Modeling in Life Insurance
SOA Predictive Analytics Seminar Hong Kong 29 Aug. 2018 Hong Kong Session 5 Predictive Modeling in Life Insurance Jingyi Zhang, Ph.D Predictive Modeling in Life Insurance JINGYI ZHANG PhD Scientist Global
More informationCHAPTER 8 EXAMPLES: MIXTURE MODELING WITH LONGITUDINAL DATA
Examples: Mixture Modeling With Longitudinal Data CHAPTER 8 EXAMPLES: MIXTURE MODELING WITH LONGITUDINAL DATA Mixture modeling refers to modeling with categorical latent variables that represent subpopulations
More informationKeywords Akiake Information criterion, Automobile, Bonus-Malus, Exponential family, Linear regression, Residuals, Scaled deviance. I.
Application of the Generalized Linear Models in Actuarial Framework BY MURWAN H. M. A. SIDDIG School of Mathematics, Faculty of Engineering Physical Science, The University of Manchester, Oxford Road,
More informationMultiple Regression and Logistic Regression II. Dajiang 525 Apr
Multiple Regression and Logistic Regression II Dajiang Liu @PHS 525 Apr-19-2016 Materials from Last Time Multiple regression model: Include multiple predictors in the model = + + + + How to interpret the
More informationLondon Market Pricing Framework
London Market Pricing Framework Hannes van Rensburg, Watson Wyatt Ryan Warren, Watson Wyatt GIRO 2009 - Edinburgh 8 October 2009 1 London Market Pricing Framework What we will cover Pricing framework Overview
More informationNegative Binomial Model for Count Data Log-linear Models for Contingency Tables - Introduction
Negative Binomial Model for Count Data Log-linear Models for Contingency Tables - Introduction Statistics 149 Spring 2006 Copyright 2006 by Mark E. Irwin Negative Binomial Family Example: Absenteeism from
More informationSTA 4504/5503 Sample questions for exam True-False questions.
STA 4504/5503 Sample questions for exam 2 1. True-False questions. (a) For General Social Survey data on Y = political ideology (categories liberal, moderate, conservative), X 1 = gender (1 = female, 0
More informationIntro to GLM Day 2: GLM and Maximum Likelihood
Intro to GLM Day 2: GLM and Maximum Likelihood Federico Vegetti Central European University ECPR Summer School in Methods and Techniques 1 / 32 Generalized Linear Modeling 3 steps of GLM 1. Specify the
More informationLog-linear Modeling Under Generalized Inverse Sampling Scheme
Log-linear Modeling Under Generalized Inverse Sampling Scheme Soumi Lahiri (1) and Sunil Dhar (2) (1) Department of Mathematical Sciences New Jersey Institute of Technology University Heights, Newark,
More informationGamma Distribution Fitting
Chapter 552 Gamma Distribution Fitting Introduction This module fits the gamma probability distributions to a complete or censored set of individual or grouped data values. It outputs various statistics
More informationStochastic Analysis Of Long Term Multiple-Decrement Contracts
Stochastic Analysis Of Long Term Multiple-Decrement Contracts Matthew Clark, FSA, MAAA and Chad Runchey, FSA, MAAA Ernst & Young LLP January 2008 Table of Contents Executive Summary...3 Introduction...6
More information[D7] PROBABILITY DISTRIBUTION OF OUTSTANDING LIABILITY FROM INDIVIDUAL PAYMENTS DATA Contributed by T S Wright
Faculty and Institute of Actuaries Claims Reserving Manual v.2 (09/1997) Section D7 [D7] PROBABILITY DISTRIBUTION OF OUTSTANDING LIABILITY FROM INDIVIDUAL PAYMENTS DATA Contributed by T S Wright 1. Introduction
More informationBy-Peril Deductible Factors
By-Peril Deductible Factors Luyang Fu, Ph.D., FCAS Jerry Han, Ph.D., ASA March 17 th 2010 State Auto is one of only 13 companies to earn an A+ Rating by AM Best every year since 1954! Agenda Introduction
More informationHierarchical Generalized Linear Models. Measurement Incorporated Hierarchical Linear Models Workshop
Hierarchical Generalized Linear Models Measurement Incorporated Hierarchical Linear Models Workshop Hierarchical Generalized Linear Models So now we are moving on to the more advanced type topics. To begin
More informationContents Utility theory and insurance The individual risk model Collective risk models
Contents There are 10 11 stars in the galaxy. That used to be a huge number. But it s only a hundred billion. It s less than the national deficit! We used to call them astronomical numbers. Now we should
More informationAPPROACHES TO VALIDATING METHODOLOGIES AND MODELS WITH INSURANCE APPLICATIONS
APPROACHES TO VALIDATING METHODOLOGIES AND MODELS WITH INSURANCE APPLICATIONS LIN A XU, VICTOR DE LA PAN A, SHAUN WANG 2017 Advances in Predictive Analytics December 1 2, 2017 AGENDA QCRM to Certify VaR
More informationSTAT 157 HW1 Solutions
STAT 157 HW1 Solutions http://www.stat.ucla.edu/~dinov/courses_students.dir/10/spring/stats157.dir/ Problem 1. 1.a: (6 points) Determine the Relative Frequency and the Cumulative Relative Frequency (fill
More informationMeasuring Policyholder Behavior in Variable Annuity Contracts
Insights September 2010 Measuring Policyholder Behavior in Variable Annuity Contracts Is Predictive Modeling the Answer? by David J. Weinsier and Guillaume Briere-Giroux Life insurers that write variable
More informationLoss Simulation Model Testing and Enhancement
Loss Simulation Model Testing and Enhancement Casualty Loss Reserve Seminar By Kailan Shang Sept. 2011 Agenda Research Overview Model Testing Real Data Model Enhancement Further Development Enterprise
More informationproc genmod; model malform/total = alcohol / dist=bin link=identity obstats; title 'Table 2.7'; title2 'Identity Link';
BIOS 6244 Analysis of Categorical Data Assignment 5 s 1. Consider Exercise 4.4, p. 98. (i) Write the SAS code, including the DATA step, to fit the linear probability model and the logit model to the data
More informationAn Actuarial Evaluation of the Insurance Limits Buying Decision
An Actuarial Evaluation of the Insurance Limits Buying Decision Joe Wieligman Client Executive VP Hylant Travis J. Grulkowski Principal & Consulting Actuary Milliman, Inc. WWW.CHICAGOLANDRISKFORUM.ORG
More informationthe display, exploration and transformation of the data are demonstrated and biases typically encountered are highlighted.
1 Insurance data Generalized linear modeling is a methodology for modeling relationships between variables. It generalizes the classical normal linear model, by relaxing some of its restrictive assumptions,
More informationBayesian Multinomial Model for Ordinal Data
Bayesian Multinomial Model for Ordinal Data Overview This example illustrates how to fit a Bayesian multinomial model by using the built-in mutinomial density function (MULTINOM) in the MCMC procedure
More informationLecture 21: Logit Models for Multinomial Responses Continued
Lecture 21: Logit Models for Multinomial Responses Continued Dipankar Bandyopadhyay, Ph.D. BMTRY 711: Analysis of Categorical Data Spring 2011 Division of Biostatistics and Epidemiology Medical University
More informationPricing Analytics for the Small and Medium Sized Company
Pricing Analytics for the Small and Medium Sized Company The Road to Advanced Pricing Practices 2014 CAS RPM By: Len Llaguno April 1, 2014 2014 Towers Watson. All rights reserved. 0 Antitrust Notice The
More informationConverting to the Standard Normal rv: Exponential PDF and CDF for x 0 Chapter 7: expected value of x
Key Formula Sheet ASU ECN 22 ASWCC Chapter : no key formulas Chapter 2: Relative Frequency=freq of the class/n Approx Class Width: =(largest value-smallest value) /number of classes Chapter 3: sample and
More information9. Logit and Probit Models For Dichotomous Data
Sociology 740 John Fox Lecture Notes 9. Logit and Probit Models For Dichotomous Data Copyright 2014 by John Fox Logit and Probit Models for Dichotomous Responses 1 1. Goals: I To show how models similar
More informationMaster s in Financial Engineering Foundations of Buy-Side Finance: Quantitative Risk and Portfolio Management. > Teaching > Courses
Master s in Financial Engineering Foundations of Buy-Side Finance: Quantitative Risk and Portfolio Management www.symmys.com > Teaching > Courses Spring 2008, Monday 7:10 pm 9:30 pm, Room 303 Attilio Meucci
More informationCambridge University Press Risk Modelling in General Insurance: From Principles to Practice Roger J. Gray and Susan M.
adjustment coefficient, 272 and Cramér Lundberg approximation, 302 existence, 279 and Lundberg s inequality, 272 numerical methods for, 303 properties, 272 and reinsurance (case study), 348 statistical
More informationContents Part I Descriptive Statistics 1 Introduction and Framework Population, Sample, and Observations Variables Quali
Part I Descriptive Statistics 1 Introduction and Framework... 3 1.1 Population, Sample, and Observations... 3 1.2 Variables.... 4 1.2.1 Qualitative and Quantitative Variables.... 5 1.2.2 Discrete and Continuous
More informationKey Features Asset allocation, cash flow analysis, object-oriented portfolio optimization, and risk analysis
Financial Toolbox Analyze financial data and develop financial algorithms Financial Toolbox provides functions for mathematical modeling and statistical analysis of financial data. You can optimize portfolios
More informationDuangporn Jearkpaporn, Connie M. Borror Douglas C. Montgomery and George C. Runger Arizona State University Tempe, AZ
Process Monitoring for Correlated Gamma Distributed Data Using Generalized Linear Model Based Control Charts Duangporn Jearkpaporn, Connie M. Borror Douglas C. Montgomery and George C. Runger Arizona State
More informationA case study on using generalized additive models to fit credit rating scores
Int. Statistical Inst.: Proc. 58th World Statistical Congress, 2011, Dublin (Session CPS071) p.5683 A case study on using generalized additive models to fit credit rating scores Müller, Marlene Beuth University
More information11. Logistic modeling of proportions
11. Logistic modeling of proportions Retrieve the data File on main menu Open worksheet C:\talks\strirling\employ.ws = Note Postcode is neighbourhood in Glasgow Cell is element of the table for each postcode
More informationHow Advanced Pricing Analysis Can Support Underwriting by Claudine Modlin, FCAS, MAAA
How Advanced Pricing Analysis Can Support Underwriting by Claudine Modlin, FCAS, MAAA September 21, 2014 2014 Towers Watson. All rights reserved. 3 What Is Predictive Modeling Predictive modeling uses
More informationOutline. Review Continuation of exercises from last time
Bayesian Models II Outline Review Continuation of exercises from last time 2 Review of terms from last time Probability density function aka pdf or density Likelihood function aka likelihood Conditional
More informationPRE CONFERENCE WORKSHOP 3
PRE CONFERENCE WORKSHOP 3 Stress testing operational risk for capital planning and capital adequacy PART 2: Monday, March 18th, 2013, New York Presenter: Alexander Cavallo, NORTHERN TRUST 1 Disclaimer
More informationMAKING CLAIMS APPLICATIONS OF PREDICTIVE ANALYTICS IN LONG-TERM CARE BY ROBERT EATON AND MISSY GORDON
MAKING CLAIMS APPLICATIONS OF PREDICTIVE ANALYTICS IN LONG-TERM CARE BY ROBERT EATON AND MISSY GORDON Predictive analytics has taken far too long in getting its foothold in the long-term care (LTC) insurance
More informationSTK Lecture 7 finalizing clam size modelling and starting on pricing
STK 4540 Lecture 7 finalizing clam size modelling and starting on pricing Overview Important issues Models treated Curriculum Duration (in lectures) What is driving the result of a nonlife insurance company?
More informationStatistical Analysis of Life Insurance Policy Termination and Survivorship
Statistical Analysis of Life Insurance Policy Termination and Survivorship Emiliano A. Valdez, PhD, FSA Michigan State University joint work with J. Vadiveloo and U. Dias Sunway University, Malaysia Kuala
More informationCHAPTER 12 EXAMPLES: MONTE CARLO SIMULATION STUDIES
Examples: Monte Carlo Simulation Studies CHAPTER 12 EXAMPLES: MONTE CARLO SIMULATION STUDIES Monte Carlo simulation studies are often used for methodological investigations of the performance of statistical
More informationIllustration 1: Determinants of Firm Debt
Illustration 1: Determinants of Firm Debt Consider the file CentralBalancos-BP.dta, which comprises accounting data for Portuguese firms. The aim is to explain the proportion of debt in the firm s capital
More informationAdvanced Risk Management Use of Predictive Modeling in Underwriting and Pricing
Advanced Risk Management Use of Predictive Modeling in Underwriting and Pricing By Saikat Maitra & Debashish Banerjee Abstract In this paper, the authors describe data mining and predictive modeling techniques
More informationOrdinal Multinomial Logistic Regression. Thom M. Suhy Southern Methodist University May14th, 2013
Ordinal Multinomial Logistic Thom M. Suhy Southern Methodist University May14th, 2013 GLM Generalized Linear Model (GLM) Framework for statistical analysis (Gelman and Hill, 2007, p. 135) Linear Continuous
More informationNon linearity issues in PD modelling. Amrita Juhi Lucas Klinkers
Non linearity issues in PD modelling Amrita Juhi Lucas Klinkers May 2017 Content Introduction Identifying non-linearity Causes of non-linearity Performance 2 Content Introduction Identifying non-linearity
More informationCopyright 2011 Pearson Education, Inc. Publishing as Addison-Wesley.
Appendix: Statistics in Action Part I Financial Time Series 1. These data show the effects of stock splits. If you investigate further, you ll find that most of these splits (such as in May 1970) are 3-for-1
More informationIntroduction to General and Generalized Linear Models
Introduction to General and Generalized Linear Models Generalized Linear Models - IIIb Henrik Madsen March 18, 2012 Henrik Madsen () Chapman & Hall March 18, 2012 1 / 32 Examples Overdispersion and Offset!
More informationStochastic Claims Reserving _ Methods in Insurance
Stochastic Claims Reserving _ Methods in Insurance and John Wiley & Sons, Ltd ! Contents Preface Acknowledgement, xiii r xi» J.. '..- 1 Introduction and Notation : :.... 1 1.1 Claims process.:.-.. : 1
More informationBuilding and Checking Survival Models
Building and Checking Survival Models David M. Rocke May 23, 2017 David M. Rocke Building and Checking Survival Models May 23, 2017 1 / 53 hodg Lymphoma Data Set from KMsurv This data set consists of information
More informationAnalytics on pension valuations
Analytics on pension valuations Research Paper Business Analytics Author: Arno Hendriksen November 4, 2017 Abstract EY Actuaries performs pension calculations for several companies where both the the assets
More informationGeneralized Linear Models
Generalized Linear Models Scott Creel Wednesday, September 10, 2014 This exercise extends the prior material on using the lm() function to fit an OLS regression and test hypotheses about effects on a parameter.
More informationActuarial Society of India EXAMINATIONS
Actuarial Society of India EXAMINATIONS 7 th June 005 Subject CT6 Statistical Models Time allowed: Three Hours (0.30 am 3.30 pm) INSTRUCTIONS TO THE CANDIDATES. Do not write your name anywhere on the answer
More information1. You are given the following information about a stationary AR(2) model:
Fall 2003 Society of Actuaries **BEGINNING OF EXAMINATION** 1. You are given the following information about a stationary AR(2) model: (i) ρ 1 = 05. (ii) ρ 2 = 01. Determine φ 2. (A) 0.2 (B) 0.1 (C) 0.4
More information*9-BES2_Logistic Regression - Social Economics & Public Policies Marcelo Neri
Econometric Techniques and Estimated Models *9 (continues in the website) This text details the different statistical techniques used in the analysis, such as logistic regression, applied to discrete variables
More informationTherefore, statistical modelling tools are required which make thorough space-time analyses of insurance regression data possible and allow to explore
Bayesian space time analysis of health insurance data Stefan Lang, Petra Kragler, Gerhard Haybach and Ludwig Fahrmeir University of Munich, Ludwigstr. 33, 80539 Munich email: lang@stat.uni-muenchen.de
More informationStatistics and Finance
David Ruppert Statistics and Finance An Introduction Springer Notation... xxi 1 Introduction... 1 1.1 References... 5 2 Probability and Statistical Models... 7 2.1 Introduction... 7 2.2 Axioms of Probability...
More informationModel fit assessment via marginal model plots
The Stata Journal (2010) 10, Number 2, pp. 215 225 Model fit assessment via marginal model plots Charles Lindsey Texas A & M University Department of Statistics College Station, TX lindseyc@stat.tamu.edu
More informationHomework Problems Stat 479
Chapter 10 91. * A random sample, X1, X2,, Xn, is drawn from a distribution with a mean of 2/3 and a variance of 1/18. ˆ = (X1 + X2 + + Xn)/(n-1) is the estimator of the distribution mean θ. Find MSE(
More informationLab 2 - Decision theory
Lab 2 - Decision theory Edvin Listo Zec 920625-2976 edvinli@student.chalmers.se September 29, 2014 Co-worker: Jessica Fredby Introduction The goal of this computer assignment is to analyse a given set
More informationOperational Risk Aggregation
Operational Risk Aggregation Professor Carol Alexander Chair of Risk Management and Director of Research, ISMA Centre, University of Reading, UK. Loss model approaches are currently a focus of operational
More informationMaximum Likelihood Estimation
Maximum Likelihood Estimation EPSY 905: Fundamentals of Multivariate Modeling Online Lecture #6 EPSY 905: Maximum Likelihood In This Lecture The basics of maximum likelihood estimation Ø The engine that
More informationCase Study: Heavy-Tailed Distribution and Reinsurance Rate-making
Case Study: Heavy-Tailed Distribution and Reinsurance Rate-making May 30, 2016 The purpose of this case study is to give a brief introduction to a heavy-tailed distribution and its distinct behaviors in
More informationModeling. joint work with Jed Frees, U of Wisconsin - Madison. Travelers PASG (Predictive Analytics Study Group) Seminar Tuesday, 12 April 2016
joint work with Jed Frees, U of Wisconsin - Madison Travelers PASG (Predictive Analytics Study Group) Seminar Tuesday, 12 April 2016 claim Department of Mathematics University of Connecticut Storrs, Connecticut
More informationWesVar uses repeated replication variance estimation methods exclusively and as a result does not offer the Taylor Series Linearization approach.
CHAPTER 9 ANALYSIS EXAMPLES REPLICATION WesVar 4.3 GENERAL NOTES ABOUT ANALYSIS EXAMPLES REPLICATION These examples are intended to provide guidance on how to use the commands/procedures for analysis of
More informationINSTITUTE OF ACTUARIES OF INDIA EXAMINATIONS. 20 th May Subject CT3 Probability & Mathematical Statistics
INSTITUTE OF ACTUARIES OF INDIA EXAMINATIONS 20 th May 2013 Subject CT3 Probability & Mathematical Statistics Time allowed: Three Hours (10.00 13.00) Total Marks: 100 INSTRUCTIONS TO THE CANDIDATES 1.
More informationBonus-malus systems 6.1 INTRODUCTION
6 Bonus-malus systems 6.1 INTRODUCTION This chapter deals with the theory behind bonus-malus methods for automobile insurance. This is an important branch of non-life insurance, in many countries even
More information**BEGINNING OF EXAMINATION** A random sample of five observations from a population is:
**BEGINNING OF EXAMINATION** 1. You are given: (i) A random sample of five observations from a population is: 0.2 0.7 0.9 1.1 1.3 (ii) You use the Kolmogorov-Smirnov test for testing the null hypothesis,
More informationData Mining Applications in Health Insurance
Data Mining Applications in Health Insurance Salford Systems Data Mining Conference New York, NY March 28-30, 2005 Lijia Guo,, PhD, ASA, MAAA University of Central Florida 1 Agenda Introductions to Data
More informationCS 361: Probability & Statistics
March 12, 2018 CS 361: Probability & Statistics Inference Binomial likelihood: Example Suppose we have a coin with an unknown probability of heads. We flip the coin 10 times and observe 2 heads. What can
More informationReview questions for Multinomial Logit/Probit, Tobit, Heckit, Quantile Regressions
1. I estimated a multinomial logit model of employment behavior using data from the 2006 Current Population Survey. The three possible outcomes for a person are employed (outcome=1), unemployed (outcome=2)
More informationSession 55 PD, Individual Life Mortality Experience Study Results. Moderator: Cynthia MacDonald, FSA, CFA, MAAA
SOA Antitrust Disclaimer SOA Presentation Disclaimer Session 55 PD, Individual Life Mortality Experience Study Results Moderator: Cynthia MacDonald, FSA, CFA, MAAA Presenters: Roland Fawthrop, FSA, MAAA
More informationFinancial Analysis The Price of Risk. Skema Business School. Portfolio Management 1.
Financial Analysis The Price of Risk bertrand.groslambert@skema.edu Skema Business School Portfolio Management Course Outline Introduction (lecture ) Presentation of portfolio management Chap.2,3,5 Introduction
More informationM249 Diagnostic Quiz
THE OPEN UNIVERSITY Faculty of Mathematics and Computing M249 Diagnostic Quiz Prepared by the Course Team [Press to begin] c 2005, 2006 The Open University Last Revision Date: May 19, 2006 Version 4.2
More informationMaximum Likelihood Estimation
Maximum Likelihood Estimation The likelihood and log-likelihood functions are the basis for deriving estimators for parameters, given data. While the shapes of these two functions are different, they have
More informationInternet Appendix for. On the High Frequency Dynamics of Hedge Fund Risk Exposures
Internet Appendix for On the High Frequency Dynamics of Hedge Fund Risk Exposures This internet appendix provides supplemental analyses to the main tables in On the High Frequency Dynamics of Hedge Fund
More informationCompetition price analysis in non-life insurance
White Paper on Non-Life Insurance: Competition A Reacfin price White analysis Paper in on non-life Non-Life insurance Insurance: - How machine learning and statistical predictive models can help Competition
More informationLongitudinal Logistic Regression: Breastfeeding of Nepalese Children
Longitudinal Logistic Regression: Breastfeeding of Nepalese Children Scientific Question Determine whether the breastfeeding of Nepalese children varies with child age and/or sex of child. Data: Nepal
More information4-2 Probability Distributions and Probability Density Functions. Figure 4-2 Probability determined from the area under f(x).
4-2 Probability Distributions and Probability Density Functions Figure 4-2 Probability determined from the area under f(x). 4-2 Probability Distributions and Probability Density Functions Definition 4-2
More informationExam STAM Practice Exam #1
!!!! Exam STAM Practice Exam #1 These practice exams should be used during the month prior to your exam. This practice exam contains 20 questions, of equal value, corresponding to about a 2 hour exam.
More informationStat 328, Summer 2005
Stat 328, Summer 2005 Exam #2, 6/18/05 Name (print) UnivID I have neither given nor received any unauthorized aid in completing this exam. Signed Answer each question completely showing your work where
More informationSTATISTICAL DISTRIBUTIONS AND THE CALCULATOR
STATISTICAL DISTRIBUTIONS AND THE CALCULATOR 1. Basic data sets a. Measures of Center - Mean ( ): average of all values. Characteristic: non-resistant is affected by skew and outliers. - Median: Either
More informationI L L I N O I S UNIVERSITY OF ILLINOIS AT URBANA-CHAMPAIGN
Modeling Counts & ZIP: Extended Example Carolyn J. Anderson Department of Educational Psychology I L L I N O I S UNIVERSITY OF ILLINOIS AT URBANA-CHAMPAIGN Modeling Counts Slide 1 of 36 Outline Outline
More informationCalculating the Probabilities of Member Engagement
Calculating the Probabilities of Member Engagement by Larry J. Seibert, Ph.D. Binary logistic regression is a regression technique that is used to calculate the probability of an outcome when there are
More informationModelling the potential human capital on the labor market using logistic regression in R
Modelling the potential human capital on the labor market using logistic regression in R Ana-Maria Ciuhu (dobre.anamaria@hotmail.com) Institute of National Economy, Romanian Academy; National Institute
More information9/5/2013. An Approach to Modeling Pharmaceutical Liability. Casualty Loss Reserve Seminar Boston, MA September Overview.
An Approach to Modeling Pharmaceutical Liability Casualty Loss Reserve Seminar Boston, MA September 2013 Overview Introduction Background Model Inputs / Outputs Model Mechanics Q&A Introduction Business
More information