• • • • • Introduction A recursive approach A Gerber Shiu function at claim instants Numerical illustrations Conclusions • Chan et al. (2003) , Dang et al. (2009) U i (t ) ui ci t Ni (t ) i X k, t 0; i 1,2 (1) k 1 • u i - the initial capital of the i-th class of business; • ci - the premium rate of the i-th class of business; i • X k - the k-th claim amount in the i-th risk process, with common cdf Fi () and pdf f i () ; • N i (t ) - the counting process for the i-th risk process. N1 (t ) and N 2(t ) are common shock correlated Poisson processes occurring at rates 1 and 2 respectively. N1 (t ) N11 (t ) N12 (t ) N2 (t ) N22 (t ) N12 (t ) where N11 (t ), N22 (t ) and N12 (t ) are independent Poisson processes with rates 11 , 22 and 12 ; • Tor inf{t 0 | min{U1 (t ),U 2 (t )} 0} min(1, 2 ) • Tsim inf{t 0 | max{U1 (t ),U 2 (t )} 0} • Tand max(1 , 2 ) • Tsum inf{t 0 | U1 (t ) U 2 (t ) 0} • Chan et al. (2003) • Cai and Li (2005) • Yuen et al. (2006) • Li et al. (2007) • Dang et al. (2009) Chan et al. (2003) for 11 22 0 • c1 • u1 u2 (u1 , u2 ) (u1 , u2 ) c2 12 (u1 , u2 ) 12 (u1 z1 , u2 z2 )dF2 ( z2 )dF1 ( z1 ) (2) 0 0 u1 u2 Dang et al. (2009) u 12 1 2 n 1 (u1 , u2 ) { 12 c11 c2 2 0 1 u1 u2 c2 ( a1 u1 ) u 2 c1 0 c1 ( a2 u 2 ) u1 c2 0 u2 0 n (a1 , a2 )e 1 ( a1 u1 ) e 2 ( a2 u2 ) da2 da1 12 n (a1 , a2 )e c1 ( u1 a1 ) 2 [ 12 n (a1 , a2 )e c2 wit h st art ingpoint 0 (u1 , u2 ) 1 e c2 ( a1 u1 ) u 2 a2 ] c1 ( u 2 a2 ) 1 [ e da2 da1 c1 ( a2 u 2 ) u1 a1 ] c2 da1 da2 (3) 0 (u1 , u 2 ) 1, 1 (u1 , u 2 ) 0 0 u 2 c2t 0 0 u1 c1t 0 0 0 u 2 c2t u 2 c2t 0 0 f1 ( x1 )11e ( 11 22 12 ) t dx1dt 0 u 2 c2t 0 f 2 ( x2 )22 e ( 11 22 12 ) t dx2 dt f1 ( x1 ) f 2 ( x2 )12 e ( 11 22 12 ) t dx1dx2 dt, 0 n 1 (u1 , u 2 ) u1 c1t u1 c1t 0 n (u1 c1t x1 , u 2 c2t ) f1 ( x1 )11e ( 11 22 12 ) t dx1dt (4) n (u1 c1t , u 2 c2t x2 ) f 2 ( x2 )22 e ( 11 22 12 ) t dx2 dt u1 c1t 0 n 1 (u1 , u 2 ) n (u1 c1t x1 , u 2 c2t x2 ) f1 ( x1 ) f 2 ( x2 )12 e ( 11 22 12 ) t dx1dx2 dt 12 1 2 S c11 c2 2 12 1 2 22 2 c1 S c11 c2 2 u1 u2 0 0 n ( a1 , a2 )e 1 ( a1 u1 ) e 2 ( a2 u2 ) da2 da1 s ( u1 a1 ) 2 [ 2 ( a1 u1 ) u 2 a2 ] cc2 ( a1 u1 ) u 2 c1 c1 1 n ( a1 , a2 )e e da2 da1 u1 0 c s ( u 2 a2 ) 1 [ 1 ( a 2 u 2 ) u1 a1 ] 12 1 2 11 1 cc12 ( a2 u2 ) u1 c2 n ( a1 , a2 )e e c2 da1da2 u2 0 c c c 1 1 2 2 2 S where s 11 12 22 c (5) • Let {X k1}k 1 and{X k2}k 1 follow independent PH distributions with parameters (α, T) and (β, Q). n 1 (u1 , u2 ) u1 u 2 0 0 n (a1 , a2 )12 ( )[12 (c1T ) (c2Q)]1 e[T (u1 a1 )][Q (u2 a2 )] (t q)da2 da1 u1 c2 ( a1 u1 ) u 2 c1 0 n (a1 , a2 )12 ( )[12 (c1T ) (c2Q)]1 e [ 12 ( c1T ) ( c2Q )]( a1 u1 ) c1 e[T (u1 a1 )][Q (u2 a2 )] (t q)da2 da1 u2 c1 ( a2 u 2 ) u1 c2 0 n (a1 , a2 )12 ( )[12 (c1T ) (c2Q)]1 e e[T (u1 a1 )][Q (u2 a2 )] (t q)da1da2 [ 12 ( c1T ) ( c2Q )]( a2 u 2 ) c2 (6) • m(u1, u2 ) E(u1 ,u2 ) [eTo r w(U1 (Tor ),U2 (Tor ))I (Tor )] (7) • w(,) is a penalty function that depends on the surplus levels at time Tor in both processes. w(U1 (Tor ),U 2 (Tor )) w1 (U1 (Tor ),U 2 (Tor ))I ( 1 2 ) w2 (U1 (Tor ),U 2 (Tor ))I ( 2 1 ) w12 (U1 (Tor ),U 2 (Tor ))I ( 1 2 ) (8) • Here are few choices of the penalty functions 1. w1 (,) w2 (,) w12 (,) 1 2. 0, w1 (,) w2 (,) 0 and w12 (,) 1 3. w1 ( y, z) y, w2 ( y, z) z and w12 ( y, z) y z E(u1 ,u2 ) [e1 | U1 (1 ) | I (1 2 ;1 )] E(u1 ,u2 ) [e 2 | U2 ( 2 ) | I ( 2 1; 2 )] (9) 4. w1 ( y, z) y z and w2 (,) w12 (,) 0 E(u ,u ) [e (U1 (1 ) U2 ( 2 ))I (1 2 )] (10) 1 1 2 • mn (u1, u2 ) E(u1 ,u2 ) [e • m(u1 , u2 ) To r w(U1 (Tor ),U2 (Tor ))I (Tor Sn )] (11) n 1 mn (u1 , u2 ) (12) • m1 (u1 , u2 ) m1 (u1 , u2 ) m1 (u1 , u2 ) m1 1 2 12 (u1 , u2 ) (13) Where m1 (u1 , u2 ), m1 (u1 , u2 ) and m1 (u1 , u2 ) correspond to the cases {τ1<τ2}, {τ2<τ1} and {τ1=τ2} respectively. 1 2 12 • Considering the first case when ruin occurs at the first claim instant in {U1(t)} only and using a conditional argument gives m (u1 , u2 ) 1 1 0 0 u 2 c2 t 0 0 0 w1 ( y, u2 c2t ) f1 (u1 c1t y )11e ( s )t dydt (14) w1 ( y, z ) f1 (u1 c1t y ) f 2 (u2 c2t z )12 e ( s )t dzdydt • By similar method, one immediately has m12 and m112. Hence by adding m11 , m12 and m112, we obtain the starting point of recursion. • If w1 ( y, z) y, w2 ( y, z) z , and w12 y z , the three integrals reduce to m11 (u1 , u2 ) 0 m12 (u1 , u2 ) 0 0 m (u1 , u2 ) 12 1 y f1 (u1 c1t y )[11 12 F2 (u2 c2t )]e ( s ) t dydt, 0 0 0 y f 2 (u2 c2t y )[22 12 F1 (u1 c1t )]e ( s ) t dydt, y u1 c1t y u1 c1t (15) f1 ( x1 ) f 2 (u2 c2t u1 c1t y x1 )12 e ( s ) t dx1dydt. • mn1 (u1 , u2 ) 0 0 0 u 2 c2 t u 2 c2 t 0 0 u1 c1t 0 mn (u1 c1t x1 , u2 c2t ) f1 ( x1 )11e ( s ) dx1dt mn (u1 c1t , u2 c2t x2 ) f 2 ( x2 )22 e ( s ) dx2 dt u1 c1t 0 (16) mn (u1 c1t x1 , u2 c2t x2 ) f1 ( x1 ) f 2 ( x2 )12 e ( s ) dx1dx2 dt • The idea that we use to find a computational tractable solution of (16) is based on mathematical induction. 11 12 12 • m11 (u1 , u2 ) e u e ( u u ) , 1 1 1 1 2 2 1 (s c11 ) 1 (s c11 c2 2 ) 22 12 12 m12 (u1 , u 2 ) e u e ( u u ) , 2 ( s c2 2 ) 2 (s c11 c2 2 ) 1 1 12 m112 (u1 , u2 ) ( ) e ( u u ) . 1 2 s c11 c2 2 2 2 1 1 1 1 2 2 (17) 2 2 • Therefore, the expected discounted deficit when ruin happens at the instant of the first claim is given by 11 12 22 12 m1 (u1 , u2 ) e u e u (18) 1 (s c11 ) 2 (s c2 2 ) 1 1 2 2 mn 1 (u1 , u 2 ) n n n a[ n 1, j ]u1j e 1u1 b[ n 1, j ]u2j e 2u2 j 0 j 0 k 0 n e j 0 u1j u 2k e ( 1u1 2u 2 ) , [ n 1, j , k ] for n 0,1, , wit h a[ n 1, j ] b[ n 1, j ] a[1, 0 ] 12 a[ n ,i ]i!c1i 1 j n 1 i max( j 1, 0 ) j!(s c1 1 ) i 1 j b[1, 0 ] b[ n ,i ]i!c n 1 i 1 j 2 i 1 j 2 2 2 i max( j 1, 0 ) j!(s c2 ) n 1 e[ n 1, j , k ] I ( k 0)( i max( j 1, 0 ) I ( j 0)( n 1 i j n 1 i j 22 a[ n ,i ]i!c1i j , j!(s c1 1 ) i 1 j 11b[ n ,i ]i!c2i j , j!(s c2 2 ) i 1 j n 1 1 a[ n ,i ]c1i 1 j i! 22 a[ n ,i ]c1i j i! ) i 1 j j!(s c11 c2 2 ) i 1 j i j j!(s c1 1 c2 2 ) n 1 i max(k 1, 0 ) n 1 2 b[ n ,i ]c2i 1 k i! 11b[ n ,i ]c2i k i! ) i 1 k k!(s c1 1 c2 2 ) i 1 k i j k!(s c1 1 c2 2 ) e[ n ,i , q ]22 2i! q!c1i j c2q 1 k i q 1 j k j! k!( c c ) i q 2 j k i j i j q max(k 1, 0 ) s 1 1 2 2 n 1 n 1 e[ n ,i , q ]11 1i! q!c1i 1 j c2q k i q 1 j k j! k!( c c ) i q 2 j k qk q k i max( j 1, 0 ) s 1 1 2 2 n 1 n 1 e[ n ,i , q ]12 1 2i! q!c1i 1 j c2q 1 k i q 2 j k q 1 k j! k!( c c ) i q 3 j k , q max(k 1, 0 ) i max( j 1, 0 ) s 1 1 2 2 n 1 n 1 12 . T hest art ingpointis given by s c11 c2 2 22 12 ,e 0. 2 (s c2 2 ) [1, 0 , 0 ] for n 1,2, ; j , k 0,1, , n, where a[1, 0 ] 11 12 ,b 1 (s c11 ) [1, 0 ] Not e t hat weassum e i j 0 for any j k . k i i 1 x 1 • f1 ( x1 ) qi x1 e , (i 1)! i 1 m 1 1 2j x2j 1e x f 2 ( x2 ) p j ( j 1)! i 1 m 2 2 • Using equation (4) for n=0 and λ11=λ22=0 along with the trivial condition 0 (u1, u2 ) 1 , we obtain m 1 1 (u1 , u 2 ) 1 a s 0 s [1, s ] 1 u e 1u1 m 1 b[1, s ]u e s 0 s 2 2u 2 m 1 m 1 e[1, s ,v ]u1s u 2v e ( 1u1 2u 2 ) , s 0 v 0 where a[1, s ] qi c1k s 1k , k s 1 i s 1k s s!( k s )!( c1 1 ) b[1, s ] p j c2k s 2k , k s 1 ) c ( )! s k ( ! s j s 1 k s 2 2 m m i 1 j 1 p j qi c1k s c2l v 1k 2l k l s v , e[1, s ,v ] k l s v 1 s k ) c c ( ! l ! k ! v ! s i s 1k s j v 1 l v 2 2 1 1 for s, v 0,1, , m 1. m i 1 m j 1 (u1 , u 2 ) lim n n (u1 , u 2 ) n 1 (u1 , u 2 ) 1 ( n 1) m 1 a[ n 1,w]u1we 1u1 w 0 ( n 1) m 1 b[ n 1,w]u2we 2u2 ( n 1) m 1( n 1) m 1 w0 w0 e y 0 u1wu 2y e ( 1u1 2u 2 ) , [ n 1, w , y ] for n 0,1, , wit h a[ n 1, w ] a[1, w ] I (0 w m 1) m nm 1 s qi a[ n, s ] i max(w mn 1,1) s max(w i , 0 ) g 0 ( 1) s g 1i c1s i w ( s i )!s! , (i 1)! g!( s g )!w!( s g i )( c11 ) s i 1 w b[ n 1, w ] b[1, w ] I (0 w m 1) m nm 1 s p j b[ n ,s ] j max(w mn 1,1) s max(w j , 0 ) g 0 ( 1) s g 2j c2s j w ( s j )!s! , ( j 1)! g!( s g )!w!( s g j )( c2 2 ) s j 1 w e[ n 1, w, y ] e[1, w, y ] I (0 w m 1,0 y m 1) m I (0 y m 1) nm 1 s m i 1 p i max(w mn 1,1) s max(w i , 0 ) g 0 i w 1 k w j qi b[ n , s ] s i k w y s ( 1) s g c1s i w c2l y 1k 2j ( s i )! s i l w y 1 siw s g (i 1)!w! y!( s g i )( c11 c2 2 ) I (0 w m 1) m nm 1 s m i 1 p i max(w mn 1,1) s max(w i , 0 ) g 0 i w 1 k w j qi b[ n , s ] s j k w y s ( 1) s g c1k w c2s j y 1k 2j ( s j )! s g ( j 1)!w! y!( s g j )( c c ) k j s w y 1 s j y 1 1 2 2 m nm 1 s m nm 1 v p j qi e[ n , s ,v ] i max(w mn 1,1) s max(w i , 0 ) g 0 j max( y mn 1,1) v max( y j , 0 ) z 0 ( s i g )(v j z ) s i v j w y s v ( 1) s v g z c1s i w c2v j y 1i 2j ( s i )!(v j )! s g v z (i 1)!( j 1)!w! y!( c c ) s v i j w y 1 , siw 1 1 2 2 for m 1,2, , y 0,1, , ( n 1) m 1. • We denote the survival probability associated to the time of ruin Tand, by and (u1, u2 ) PTand | u1, u2 P1 or 2 | u1, u2 . and 1 2 or • n (u1, u2 ) n (u1 ) n (u2 ) n (u1, u2 ) . • (u1 ) 0 1 n1 u1 c1t 0 (u1 c1t x1 ) f1 ( x1 )(11 12 )e 1 n s t 1 0 n dx1dt (u1 c1t )22est dt T heunivariat esurvivalprobability 1n 1 (u1 ) up t o and including t he(n 1) - t h claim event sadmit s t herepresent at ion n 1 n 1 (u1 ) 1 a[1n 1, j ]u1j e 1u1 , j 0 for n 0,1, , wit h a[1n 1, j ] a[11, 0 ] I ( j 0) n 1 a[11, 0 ] 1a[1n ,i ]c1i 1 j i! i max( j 1, 0 ) j!(s c11 ) i 2 j n 1 a[1n ,i ]22 c1i j i! i j j!(s c11 ) i 1 j for n 1,2, ; j 0,1, , n. T hest art ingpointis a[11, 0 ] 11 12 . s c11 , • u1=2 ,u2=10, c1=3.2, c2=30, 1/µ1=1 and 1/µ2=10. • Case 1: Independent model — λ11=λ22=2; λ12=0. Case 2: Three-states common shock model — λ11=λ22=1.5; λ12=0.5. Case 3: Three-states common shock model — λ11=λ22=0.5; λ12=1.5. Case 4: One-state common shock model — λ11 = λ22 = 0; λ12 = 2. • Note that λ1 = λ2 = 2, and θ1 = 0.6 and θ2 = 0.5. • In Case 1, after 100 iterations we obtain a ruin probability of 0.6306428 that is very close to the exact value of 0.6318894. • max{1 (u1 ), 2 (u2 ) } or (u1 , u2 ) 1 (u1 ) 2 (u2 ) 1 (u1 ) 2 (u2 ) • Cai and Li (2005, 2007) provided simple bounds for Ψand(u1, u2) given by 1 (u1 ) 2 (u 2 ) and (u1 , u 2 ) min{ 1 (u1 ), 2 (u 2 )} • δ = 0.05 • This quantity is achieved by letting w1(y, z) = y+z and w2(.,. ) = w12(.,.) =0 • This quantity is achieved by letting w2(y, z) = y+z and w1(.,. ) = w12(.,.) =0 • Several extensions: 1. Correlated claims 2. Correlated inter-arrival times and the resulting claims 3. Renewal type risk models