This Exam P sample reference tests Law of Total Variance. This is a mixed-Poisson variance calculation. The average conditional variance is 12.50 and the variance of the conditional means is 11.25, so the total variance is 23.75 and choice E.
Let N be the annual count and let Lambda be its Poisson mean after the year type is known.
Pr(Λ=20)=0.10,Pr(Λ=15)=0.30,Pr(Λ=10)=0.60
E[N∣Λ]=Λ,Var(N∣Λ)=Λ
Model
Model
Separate ordinary Poisson variability within a year type from variability among the three conditional means.
Var(N)=E[Var(N∣Λ)]+Var(E[N∣Λ])
=E[Λ]+Var(Λ)
Compute
Compute
Calculate the first two moments of the random Poisson mean and combine the two variance components.
E[Λ]=0.10(20)+0.30(15)+0.60(10)=12.50
E[Λ2]=0.10(202)+0.30(152)+0.60(102)=167.50
Var(Λ)=167.50−12.502=11.25
Var(N)=12.50+11.25=23.75
Answer
Answer
The unconditional annual count has variance 23.75.
23.75(E)
Test the method on unseen problems
Probability Multivariate & Risk Models Practice Workbook
30 original multivariate random variables problems with complete solutions and wrong-choice analysis. Written for Exam P, not copied from the SOA set. One-time $19 PDF.
These notes identify the calculation error associated with each wrong letter when that error is reproducible.
AThe value 11.25 is Var(Lambda), the between-type component only. It omits the average Poisson variance E[Lambda]=12.50 within each type.
BThe value 12.50 is E[Lambda], which is both the unconditional mean and the within-type variance component. A mixture also varies because its conditional means differ.
CThe value 12.94 is √(167.50), the root mean square of the three Poisson means. It is neither a centered variance nor the requested count variance.
DThe value 13.42 is √(180), the square root of the raw second moment E[N squared]. The question asks for variance, so neither the square root nor the uncentered moment is appropriate.
Original practice · fully worked
Original variant: recovering latent rate variability
Conditional on a latent activity rate R, the number N of detector alerts in a window is Poisson with mean R. Monitoring data give E[N]=6 and Var(N)=15. Calculate the coefficient of variation of R.
A 0.200
B 0.400
C 0.500
D 0.600
E 1.500
Answer and calculation
Mixed-Poisson moments give E[R]=6 and Var(R)=15-6=9. Thus the latent rate has standard deviation 3 and coefficient of variation 3/6=0.5, so choice C.
Setup
Setup
Use conditional Poisson expectation to recover the mean activity rate.
E[N]=E[E[N∣R]]=E[R]=6
Model
Model
Invert the mixed-Poisson total-variance identity to isolate the between-window rate variance.
Var(N)=E[R]+Var(R)
Var(R)=15−6=9
Compute
Compute
Convert the rate variance to a standard deviation and divide by its mean.
SD(R)=9=3
CV(R)=63=0.5
Answer
Answer
The activity rate has coefficient of variation 0.500.