assignment – Bayesian stats and credibility theory

answer 1)

X: number of email messages received each day

X_bar =

X|lambda ~ P(lambda)

and prior dist : lambda ~ exp(1/mu)

i) a)
f(X/lambda) = (exp(-lambda)* lambda^x)/ x!
L(X/lambda)= product(1,n): f(X/lambda)
= (exp(-lambda)* lambda^x1)/ x1 ! * (exp(-lambda)* lambda^x2)/ x2! * ..... * (exp(-
lambda)* lambda^xn)/ xn!
L(X/lambda) = exp(-n*lambda) * lambda(sum(1,n):xi) * constant

f(lambda) = (1/mu)* exp(-lambda/mu)

= exp(-lambda/mu) * constant

f(lambda/X_bar) is proportional to exp(-n*lambda) * lambda^(sum(1,n):xi)

*exp(lambda/mu)
f(lambda/X_bar) is proportional to exp(-lambda*(n+(1/mu))) *lambda^(sum(1,n):xi)

lambda/X_bar ~ gamma((sum(1,n):xi)+1 , n+(1/mu))

b)
under quadratic loss : lambda^hat =mean of posterior distribution
= ((sum(1,n): xi) + 1)/ (n+(1/mu))

in the form of credibility estimate : Xbar*Z +(1-Z)*mu

sum(1,n):xi /n * (n / (n+(1/mu)) + mu *((1/mu)/ (n+(1/mu)))

here , Xbar = sum(1,n):xi /n

prior mean = mu
z= (n / (n+(1/mu))
1-z = ((1/mu)/ (n+(1/mu)))

so , z= (n / (n+(1/mu))

c)
sum(1,n): xi = 550
mu= 50
n= 10
 lambda^hat = ((sum(1,n): xi) + 1)/ (n+(1/mu))

lambda^hat= 54.99
ii) .λ

answer 2 )

X/lambda ~ exp(lambda)

X_bar = (X1 , X2, ... Xn)

here , n= 20 and sum(1,n):xi = 24

we know ,

alpha / lambda = 1

alpha = lambda ...(A)

and , alpha / lambda^2 = ½

from (A) ,

lambda / lambda^2 = ½

lambda = 2

so , alpha = 2

therefore ,

prior distribution : lambda ~ gamma ( 2,2 )

L(X|lambda) = product(1,20): f(x|lambda)

= lambda* exp(-lambda*x1) * lambda* exp(-lambda*x2) * ... * lambda* exp(-lambda*xn)

= lambda^20 * exp(-lambda*sum(1,20):xi)

f(lambda) = lambda* exp(-2*lambda) * constant

 f(lambda|X_bar) is proportional to lambda^20 * exp(-lambda*sum(1,20):xi) *lambda* exp(-
2*lambda)

f(lambda|X_bar) is proportional to lambda^21 * exp(-lambda*(2+ sum(1,20):xi) )

lambda|X_bar ~ gamma(22 , 26)

answer 3 )

X|lambda ~ Exp(lambda)

lambda~ gamma (alpha’ , lamda’)

L(lambda) = product(1,n):f(X|lambda)

= lambda^n * exp(-lambda*sum(1,n):xi)

f(lambda) = lambda^(alpha’ -1 ) * exp(-lambda’ * lambda)

f(lambda|X_bar) ∝ lambda^( n+alpha’ -1) * exp( -lambda*( sum(1,n): (xi) + lambda’))

lambda|X_bar ~ gamma( n+alpha’ -1 , sum(1,n): (xi) + lambda’ )

answer 4)

i)

theta ~ beta( alpha, beta)

given – mean = 0.2 , variance = 0.25^2

0.2 = alpha / (alpha +beta)

alpha + beta = 5*alpha

beta = 4*alpha ......(A)

0.25^2 = (alpha*beta) / ( (alpha+beta)^2 *( alpha +beta +1) )

0.25^2 / 0.2 = (4*alpha)/ (5*alpha *(5*alpha +1 ))

0.25^2 / 0.16 = 1/ (5*alpha +1)

5*alpha +1 = 2.56

alpha = 0.312
 putting value of alpha in equation A :

beta = 1.248

ii )

X|theta ~ bin(50, theta)

L(theta) = 50 choose 12 * theta^12 * (1-theta)^38

= theta^12 * (1-theta)^38 * constant

f(theta)= theta^(alpha – 1 ) * (1-theta)^(beta-1)

f(theta|X_bar) ∝ theta^(12+alpha-1) *(1theta)^(38+beta-1 )

∝ theta^(11 + alpha) *(1-theta)^(37+beta)

theta|X_bar ~ beta(12 + alpha , beta +38)

putting is values of alpha and beta :

theta|X_bar ~ beta(12.312 , 39.248)

E(theta|X_bar) = 12.312 / (12.312+39.248) = 0.2387

iii )

theta|X_bar ~ beta ( x+ alpha , n-x+beta)

mean =( x+ alpha)/ (n+alpha+beta)

(x/n)* (n/ n+ alpha +beta) + (alpha/alpha+beta) *(alpha +beta/ n+alpha+beta)

here ,

z = n / ( n+ alpha + beta )

iv ) a )

z= 0.9697

as z tends to1 , more credible towards self risk

b)

1 ) as SD increases , data is more variable , confidence on prior data decreases and there is more
confidence on self risk , eventually z increases .
2 ) as n increases , more credibility on self risk , z increases

c)

limiting value of z is 1 , as sigma and mu increases , this means we are more confident on self risk .

answer 6 )

X : number of claims from drivers

i)
X1 ~ P(lambda) and X2 ~ P(2lambda)
L=(( exp(-lambda) * lambda^(n1)) / n1!) * (( exp(-2lambda) * 2lambda^(n2)) / n2!)
L= exp(-3lambda) * lambda^( n1 + n2) * constant
log (L) = -3lambda + (n1+n2) *ln(lambda)
d log(L) / d lambda = -3 + (n1+n2)/lambda
putting d log(L)/d lambda = 0
(n1+n2)/lambda = 3
lambda hat = (n1 + n2) / 3
ii)
a) lambda~ Exp(v)
f(lambda) = v* exp(-v*lambda) = exp(-v*lambda) * constant
f( lambda | X_bar) ∝ exp(-lambda(v+3)) * lambda^(n1+n2)

lambda|X_bar ~ gamma ( n1 + n2 + 1 , v+3 )

under quadratic loss : lambdahat = mean of posterior distribution

= (n1 + n2 + 1) / (v+3)

credibility premium:
((n1+n2) / 3) * (3/(v+3)) + (1/v) * (v/(v+3))

so , Z = 3/(v+3)

answer 7 )

L(lambda, d) = (lambda – d)^2 + d^2

E(L(lambda,d) ) = E((lambda – d)^2 + d^2)

= E(lambda^2 + d^2 – 2lambda*d + d^2)

= E(lambda^2) + d^2 -2*d*E(lambda)

= V(lambda) +(E(lambda))^2 – 2*d*E(lambda) + 2*d^2
= alpha/ beta^2 + (alpha/beta)^2 – 2*d*(alpha/beta) + 2*d^2
E(L(lambda,d) ) = (alpha*(alpha+1))/ beta^2 – (2*d*alpha)/beta + 2*d^2

d (E(L(lambda,d)) )/ d (d) = -2*lambda/beta + 4*d = 0

4d = 2*alpha / beta

dhat = alpha / 2beta

d^2(E(L(lambda,d) )) / d(d)^2 = 4 > 0

minimised at dhat = alpha / 2beta

answer 8)

i) when prior and posterior dist belong to the same family

ii) f(x) = lambda*exp(-lambda*x)

L= lambda^n *exp(-lambda*sum(1,n):xi)

lambda~ gamma (alpha’ , lambda’)

f(lambda) = lambda^(alpha’ – 1 ) * exp(-lambda*lambda’)

f(lambda|X_bar) ∝ lambda^(alpha’ + n -1) *exp( lambda’ + sum(1,n) :xi)

lambda|X_bar G(alpha’ +n , lambda’ + sum(1,n):xi)

iii) a)
lambda ~ gamma (alpha , s )
E(1/lambda) = INT (0 , Inf ): ((1/lambda) * (s^lambda *lambda^(alpha-1)
*exp(-lambda*s) )/ (alpha-1)! )d(lambda)
= s^alpha / (alpha-1)! * INT(0,inf) : [ lambda^(alpha- 2) * exp(-s*lambda) ] *d(lambda
= ( s^alpha / (alpha-1)!)* ( (alpha-2)! / s^(alpha-1))
= (s^alpha *(alpha-2) ! )/ ((alpha-1)! *s^(alpha-1) )
E(1/lambda) = s / (alpha -1)

b)
X|lambda ~ exp(lambda)
lambda~ gamma ( alpha , s)
lamda|X_bar ~ gamma (n + alpha , s+ sum(1,n):xi)
E(1/lambda) =( s+ sum(1,n):xi ) / (n + alpha – 1)

credibility premium :
((sum(1,n)xi)/n)* (n/(n+alpha-1)) + (s/(alpha-1)) * ((alpha-1)/(n+alpha-1))

so ,
Z= n/(n+alpha-1)
 iv) S_A/ (sigma_A -1) = 3
9 / (sigma_A – 2) = 0.5^2
9 = 0.25 *( sigma_A – 2)
sigma_A = 9.5/0.25
sigma_A = 38

answer 9 )

X| theta ~ N(theta , sigma^2)

L(theta) = product(1,n) : f(X|theta)

= exp((-(x1-mu)/sigma)^2) / (sigma*sqrt(2*pie))

= exp(-1/2sigma^2)*((sum(1,n):(xi^2 + theta^2 _ 2theta*xi)) * const

= exp((-1/2sigma^2 ) *(ntheta^2 – 2theta*sum(1,n):xi)) * constant

for prior :

theta ~ N(mu , sigma2^2)

f(theta) = (1/sigma2*sqrt(2*pie)) * exp ( (1/2) * ((theta-mu)/sigma2^2)^2)

f(theta) = exp( (-1/(2*sigma2^2))*(theta – mu)^2 ) * constant

f(theta) = exp( -(1/(2*sigma2^2) ) * (theta^2 + 2*theta*mu) ) * constant

f(theta|X_bar) ∝ exp[ {(-1/(2*sigma^2)) * (n*theta^2 – 2*theta*sum(1,n):x_i)} – {(1/(2*sigma2^2)*

(theta^2 +2*theta*mu)} ]

f(theta|X_bar) ∝ ((-1/2) *[ ( (theta^2/sigma2^2) – (2*theta*mu/ sigma^2) ) + ( ((n*theta^2) /

sigma^2) – ((2*theta* sum(1,n): xi)/sigma^2) ) ]

f(theta|X_bar) ∝ exp[ (-1/2)*( (n/sigma^2)+(1/sigma2^2) ) * ( theta^2 – (2*theta*((mu/sigma2^2)+

((sum(1,n):xi)/sigma^2) )/( (1/sigma2^2) + (n/sigma^2) ) )]

mu|X_bar ~ N (( (( sum(1,n):xi )/ sigma^2 ) + (mu / sigma2^2) )/ ( (1/sigma2^2) + (n/sigma^2) ), ( 1/

( (n/sigma2^2) +(1/sigma^2) ) )

ii ) under quadratic loss : theta^hat

= ( (( sum(1,n):xi )/ sigma^2 ) + (mu / sigma2^2) )/ ( (1/sigma2^2) + (n/sigma^2) )

iii ) theta^hat= [ ((sum(1,n):xi)*sigma2^2)+ (mu*sigma^2) ] / [ (n*sigma2^2) + (sigma^2) ]

credibility premium:
 [ (n*sigma2^2) / ( (n*sigma2^2) +sigma^2 ) ] * [ (sum(1,n):xi ) / n ] + [ sigma^2 /
( (n*sigma2^2) + ( sigma^2) ) ] * (mu)

here , z= (n*sigma2^2) / ( (n*sigma2^2) +sigma^2 )

company A :

Z= (5*800) /( 5* 800+500) = 0.8889

credibility premium : ( (0.8889*439)+(0.1111 *400) )* 1.25 = 21677.115

Company B:

Z = (5*600)/(5*600+350) = 0.8955

credibility premium : (0.8955 * 356 + 0.1045*300) *1.25 = 437.685

answer 11 )

ii )

X|p ~ bin(m,p)

p~ U( 0,1 )

a) because there is no information about prior dist we will take a suitable uniform dist
L(X|p) = ( N choose m ) * p ^m * (1-p)^(N-m)
L (X|p) = p^m * (1-p)^(N-m) * constant

f(p) = 1 / (1-0) = 1

f( p | X_bar ) ∝ p^m * (1-p)^(N-m)

p|X_bar ~ beta( m+1 , N-m+1)

b) under all or nothing loss , p^hat = mode of posterior dist

p^hat = (m+1-1)/(m+1+N-m-2+1)
p^hat = m/N

answer 12 )

P(X=k) = theta^k * (1-theta) ; k = 0 ,1 , 2 ....... ; 0<theta<1

f(theta) ∝ (theta^(alpha -1 ) )* (1-theta) ^(alpha-1) ; alpha>0

i) a) L(theta) = product(i=0,k): (theta^(sum(0,k):xi) * (1-theta)^n )

 logL(theta) = (sum(0,k):xi) *ln (theta) + n* ln(1-theta)
d(logL(theta)) / d(theta) = ((sum(0,k):xi)/theta) – (n/(1-theta))

putting d(logL(theta)) / d(theta) = 0

(sum(0,k):xi ) – ((sum(0,k):xi)* theta) – n*theta = 0
sum(0,k):xi = theta*((sum(0,k):xi)+n)
theta^hat = (sum(0,k):xi) / ((sum(0,k):xi)+n)

b ) f(theta|X_bar) ∝ theta ^(sum(0,k):xi) * (1-theta)^n * theta^(alpha-1) *

(1- theta)^(alpha-1)

theta|X_bar ~ beta ( alpha + sum(0,k):xi , n + alpha )

c)

answer 14 )

X: claim amount
X|mu ~ N(mu , 50^2)
mu~ N(300,20^2)

i) P( mu < 270 ) = P( (mu-300/ 20) < (270-300/20) )

= P( Z < (270-300/20) )
= P( Z < -1.5) = P(Z > 1.5)
= 1- P( Z<= 1.5)
= 0.06681 [ from tables ]

ii) a)
mu | X_bar ~ N ( [ ((2700/50^2) + (300 / 20^2)) / ((10/50^20) + (1/20^2)) ] ,
[ (10/50^20) + (1/20^2) ] )
mu |X_bar ~ N( 281.54 , 12.40^2 )

b)
P( mu < 240 ) = P(Z < (270 – 281.54)/12.40)
= P(Z< - 0.93 ) = 1- P( Z< = 0.93 )

answer 16 )

0.015 = a / (a +b)

(a+b) = 66.67 * a

b = 65.67 * a ..... (A)

and we have ,

0.005^2 = (a*b) / ((a+b) (a+b+1))

substituting values from above ,

0.005^2 = (0.015 * 65.67*a) / (( 66.67a) *( 66.67a + 1))

0.00169 = 1 / (66.67*a + 1)

66.67a + 1 = 591.01

a = 8.86

putting value of a in (A)

b = 581.836

X| q ~ bin( 58, q )

L(X|q) = 4800 choose 58 * q ^58 * (1-q) ^ 4442

= q ^ 58 * (1-q) ^ 4442 * constant

q~ beta ( 8.86 , 581.836)

f(q) = q^7.86 * (1-q)^ 580.836

f(q|X_bar) ∝ q^(58 + 7.86) * (1-q) ^ ( 4442 + 580.836)

q|X_bar ~ beta ( 66.85 , 5023.836)

answer 18 )

i) X_bar|p ~ bin( n , p )
and p ~ beta(alpha,beta)

f(X_bar | p ) = (n choose k) * p^k * (1-p)^(n-k)

L = p^k * (1-p)^(n-k) * constant

f(p) = p^(α -1) *(1-p) ^ (beta-1)

f(p |X_bar) ∝ p^k * (1-p) ^ (n-k) *p^(alpha -1 ) * (1-p) ^(beta-1)

p|X_bar ~ beta ( k + alpha , n-k+beta)

ii) X~ beta ( alpha , beta)

1
E(1/X) = ∫ ¿ ¿ x^(a -1) * (1-x)^(b-1) * (a+b-1)! / (a! * b!) )dx
0

1
= (a+b-1)! / (a! * b!) * ∫ ¿ ¿ ^(a-2) * (1-x)^(b-1) ) dx
0
 = ((a+b-1)! / (a! * b!)) *((a-2)!*(b-1)!) / (a+b-2)!
= (a-2)/ (x+b-2)
wrrrrrrrrrrrrooooooooooooongggggggggggggggggg

answer 19 )

P(p=0.4) = 0.6 , P(p=0.75) = 0.4

X_bar | p ~ bin(6, p )

P(X=x | p) = (6 choose x) * p^x * (1-p)^(6-x)

P(X=4 | p=0.4) = (6 choose 4) * 0.4^4 * 0.6^2 = 0.1382

P(X=4 | p=0.75) = (6 choose 4) * 0.75^4 * 0.25^2 = 0.2966

P(p=0.4 | X=4) = formula = 0.4113

P(p=0.75 | X=4)= 0.5887

answer 20 )

X: no . of claims per day

X_bar | mu ~ P(mu)

and

mu ~ gamma(11,0.22)

n
L(mu) = ∏ ¿ ¿
i=0

n
L(mu) = e^(-mu*n) * mu^(∑ ( xi ) ) *constant
1

f(mu) = mu^10 * x^(-0.22*mu)

f(mu|X_bar) ∝ mu^(10+ sum(1,n):xi) *e^(-mu^(n+0.22))

mu|X_bar ~ gamma ( 10+ sum(1,n):xi , n+0.22) ~ gamma( 641 , 10.22)

 under all or nothing loss , mu^hat = mode of posterior dist
mu^hat = (641-1)/10.22 = 62.622

answer 21 )

i) Y_bar | theta ~ U(0,∞ ) , theta>0

f(theta ) = alpha* (beta^alpha) * theta^(1+alpha) , theta> beta
=0 , otherwise
L(theta) = 1/theta

f(theta) = theta^(1+alpha) * constant

f( theta | Y_bar ) ∝ (theta^ (-(1+alpha)) ) / theta , for theta > max(beta_i , y_i1)
f(theta | Y_bar ) ∝ theta^ (-(1+alpha+1) ) , for theta > max (beta_i , y_1)

ii) f(theta |Y_bar) ∝ theta (-(1+alpha+ n) , theta> max(beta_i , y1 , y2 , .... , yn)

answer 22)

X_bar | P ~ bin( 7,p)

P( p=0.5 ) = 0.8 ; P(P~ U(0.5, 1) = 0.2)

P(X= x |p) = (8 choose x) *(p^x) * ((1-p)^(8-x))

P(X= 7 |p=0.5) = (8 choose 7) *(0.5^7) * 0.5 = 8 * 0.5^8

1
P(X= 7 | P~ U(0.5 , 1) ) = ∫ ((8 choose 7)∗( p )∗(1− p))dp
7

0.5

P(0.5 |X=7) =

answer 23)

X: claim amount per annum

X_bar |theta ~ N ( theta , 200^2)

theta ~ N( 600 , 50^2)

n
L = ∏ f ( X ¯¿ theta)= (exp( (-1/80000) * ( xi -mu) ^2 )) / (sigma*√ 2 π )
1

n n
= exp( - ( ∑ (xi) + (n* theta^2 ) – 2*theta * ∑ xi )/80000)
2

1 1
 n
= exp ( -(n*theta^2 – 2* theta * ∑ xi ) / 80000) *constant
1

f( theta ) =

answer 25 )

X- No of claims in a year

X_bar | mu ~ P(mu )

mu~ gamma( 10,2)

8 8
L(mu) = ∏ ((exp(−mu)∗mu (xi )
)/xi !)= exp(-n*mu) * mu^( ∫ xi¿ ¿
i=1 i=1

f(mu) = mu^9 * exp(-2mu)

f(mu|X_bar ) ∝ mu^(9+ sum xi) * exp(-mu (2+n))

mu|X_bar ~ gamma ( sum xi + 10 , 2+n)

~ gamma (18,3)

ANSWER 26 )

i) X : no of claims per month

X_bar | lambda ~ P(lambda)
lambda~ exp( 5 )

n
L = exp ( -n*lambda) * lambda ^ ∑ xi * constant
1
f(lambda) = exp( -5*lambda) * constant

n
f(lambda|X_bar) ∝ exp( -lambda*(n+5)) * lambda^(∑ xi )
1

lambda|X_bar ~ gamma ( sum xi + 1 , n+ 5)

ii) a) Bayesian estimate under quadratic loss = mean of posterior

= (sum xi + 1 ) / ( n+ 5)
 b ) Bayesian estimate under all or nothing loss = mode of posterior dist
= sum xi / (n+5)

iii) n = 5 , sum xi = 1
lambda|X_bar ~ gamma ( 2,10) ⇒ 20 X ~ chi-squared_4

under absolute error loss = median of posterior = P(Y<= M) = P(Y > M) = 0.5
20 M = 3.357
M = 0.16

answer 27 )

X: no of claims registered per week

X_Bar | lambda ~ P(lambda)

P( lambda= 1) = 0.4 ; P(lambda = 2)

P(X=x | lambda) = ( exp(-lambda) * lambda^x) / x!

P(X=3 | lambda=1) = ( exp(-1) * 1^3) / 3! = 0.061

P(X=3 | lambda=2) = ( exp(-2) * 2^3) / 3! = 0.18

P(lambda = 1 | x=3) = formula = 0.184

P(lambda =2 | x=3) = 1- 0.184 = 0.816

Lambda 1 2
P(lambda |x=3) 0.184 0.816
cdf 0.184 1

squared error loss :mean : 1.816

zero one loss : mode : 2

absolute loss : median : 2

answer 28 )

i) X: no of claims arising each month

X_bar | lambda ~ P(lambda)

alpha’ / lambda’ = 250

alpha’ / lambda’^2 = 45

on solving these equations , we get

alpha ‘ = 1387.5
 lambda’ = 5.55

lambda~ gamma (1387.5 , 5.55 )

lambda|X_bar ~ ( ∑ xi + alpha’ , n + lambda’ )
lambda|X_bar ~ ( 11888.88 , 55.55 )

under quadratic loss = lambda^ hat = 11888.88 / 55.55

lambda^hat = 214 . 02

ii) Y : no of claims per day

Y_bar | lambda ~ P(lambda/30)

P(lambda = 230 ) = 0.2 , P(lambda = 250 ) = 0.5 , P(lambda = 270) = 0.3

P(Y= y | lambda) =( exp(-lambda/30) * (lambda/30)^ (-y) ) / y!

P(Y= 7 | lambda= 230) = 0.144
P(Y= 7 | lambda = 250)= 0.133
P(Y= 7 | lambda= 270 )= 0.117

P( lambda = 230 | Y = 7 ) = formula = 0.221

P( lambda = 250 | Y = 7 ) = 0.509
P( lambda = 270 | Y = 7 ) = 0.27

Lambda 230 250 270

P( lambda|Y=7) 0.221 0.509 0.27
Cdf = 0.221 0.73 1

under quadratic loss = lambda^hat = 250.98

answer 29)

P(L) = 1/3

P(L’) = 2/3

A= late to work by more than 20 mins

P(A|L) = P(X> 20) = exp( -20/15) = 0.263

P(A|L’) = (Y>20) = (25-20) / (25-0) = 0.2

P(L|A) = (0.33 * 0.263) / ( 0.33 * 0.263 + 0.67* 0.2)

P(L|A) = 0.393

answer 30)
X_bar | p ~ bin( n, p)

p~ beta( alpha , beta)

L= (n choose k) * ( p^k ) * ( ( 1-p) ^ ( n-k) )

L = ( p^k ) * ( ( 1-p) ^ ( n-k) )* constant

f( p ) = p ^ ( alpha -1 ) * (1-p) ^ ( beta -1 )

f(p|X_bar) ∝ p^ ( k+alpha-1) * ( 1-p) ^ (n-k+beta-1)

p|X_bar ~ beta ( k+alpha , n-k+beta )

under all or nothing loss = phat = ( k+alpha-1 ) /(n+beta + alpha -2)

credibility premium :

( ( k /n) * ( n/( n+beta+alpha-2)) ) +( (( alpha-1)/(beta+alpha-2))*( (beta+alpha-2)/(n+beta+alpha -2)) )

so, Z = n/( n+beta+alpha-2)

answer 31 )

X : heights of adult males

X_bar|mu ~ N ( mu , 15^2)

mu ~ N( 187,10^2)

i)

P(mu> 180)

P(Z > (180-187)/10) =

P( Z> -0.7) = P( Z< 0.7)

= 0.75804

ii) p|X_bar ~ N ( 66.5812/ 0.3656 , 2.7578)

p |X_bar ~ (182.136 , 2.757)

p(mu>180) = P ( Z> (180-182.136)/sqrt(2.7578))

= P(Z>-2.18)
=P(Z<1.286)

answer 32

X|theta ~ bin(n,theta)
theta ~ beta ( alpha , beta )

mean = mu , var= sigma^2

i) doubt
ii) a)
f(X_bar |theta) = (n choose d)* ( theta^d) * ( 1-theta)^(n-d)
f(theta) = theta ^(alpha -1) *(1-theta)^(beta-1) * constant
f(theta|X_bar) ∝ theta ^( d+alpha -1) *(1-theta)^(n-d+beta-1)

theta|X_bar ~ beta ( d+ alpha , n-d+beta)

b ) mean = (d+alpha) / (d+n+beta)

credibility premium :
[(d/n)*(n/(alpha+n+beta))] * [(alpha/(alpha+beta) ) * ( (d+alpha) / ( d+n+beta) ) ]

so , here
Z= n/(alpha+n+beta)

answer 33 )

X: random number calls every day

X_bar | beta ~ P( beta)

we have , prior mean = 200 and prior variance = 50^2

so , on solving , we get

beta~ gamma ( 16,0.08)

L(Beta , X_bar ) = exp( - beta) * beta^240 * constant

f( beta) = beta ^ 15 * exp( -beta*0.08)

f(beta|X_bar ) ∝ exp(-beta*0.08) * beta^(255)

Beta|X_bar ~ beta( 265, 1.08)

under quadratic loss : betahat = (266) / 1.08

= 237.037

answer 34)

X_bar|p ~ bin ( n,p)

f(p) ∝ [p(1-p)]^a
 i) f(X_bar , p ) = p^sumxi * (1-p) *(nm -sumxi) * const
f(p) = [p(1-p)]^a

f(p|X_bar )∝ p^( sumxi +a) * (1-p)^(a+mn-sumxi)

p|X_bar ~ beta ( sumxi + a+1 , mn-sumxi+a +1)

ii) L(p,X_bar) = p^sumxi * (1-p) *(nm -sumxi) * const

log(L) = sumxi *log(p) + (mn-sumxi) log(1-p)
d(logL) / d(p) = sumxi/p – (mn-sumxi)/(1-p)
on equating with zero ,
phat = sumxi /(m*n)

p|X_bar beta ( sumxi + a+1 , mn-sumxi+a +1)

L=
doubt

answer 35)

X_bar |p ~ bin( 100,p)

p~ beta ( 2,8)

f(X_bar|p) = ((100 choose 13)* (p^13) * (( 1-p)^87) ) * ( ((100+g) choose 20)* (p^20) * (( 1-p)^80+g) )

L( p, X_bar) = (p^33)*((1-p) ^ (167+g) )

f(p) = p * (1-p) ^7

f(p|X_bar) ∝ (p^34) * ((1-p) ^(174+g) )

p|X_bar ~ beta ( 35 , 175+g)

under quadratic loss : mean of posterior = p^hat

p^hat = 35/(210+g)

under all or nothing loss : mode of posterior = p^hat

p^hat = 34/(208+g)

PHAT UNDER ALL OR NOTHING LOSS ≠ UNDER QUAD LOSS

answer 36)

theta ~ U ( 0,1 )

f( X_bar |theta ) = ((10 choose 5) * theta^5 * (1-theta) ^5) * ( (20choose 11) * theta ^ 11 * (1-
theta)^9)

L(theta , X-bar) = theta ^16 * (1-theta) ^ 14 * constant

f(theta) = 1

f(theta|X_bar) ∝ theta ^16 *(1-theta) ^ 14

theta|X_bar ~ beta ( 17,15)

i) theta hat = 17/(17+15)

ii) f(theta) = (theta ^(alpha -1)) * (1-theta)^(beta-1)
= (theta*(1-theta))^(alpha-1)

f(theta|X_bar) ∝ theta^(alpha+15) *(1-theta)^(alpha+13)

theta|X_bar~ beta ( alpha + 16 , alpha + 14)

under all or nothing loss :

betahat = (alpha+15 ) /(2alpha + 28 )

answer 37)
X: no of claims

X_ bar|theta ~ N (theta, τ ^2 )

theta ~ N( mu , sigma^2)

i) post : book se dekhlo

ii) under quadratic loss : basic sa ques hai main ahi kar rahi

answer 38)

P(p = 2/6) = 1/3 ; P(p = 1/6) = 2/3

P( X=4 |p) = (10choose 4) * p^4 *(1-p)^6

P( X= 4 |p = 2/6) = (10 choose 4) * (2/6) ^4 * ( 4/6)^6 = 0.2276

P( X= 4 |p = 1/6) = (10 choose 4) * (1/6) ^4 * ( 5/6)^6= 0.0542