Appendix: Some Useful Integral Identities and Symmetry Properties of Normal Random Variables
Throughout all the formulas below, a, b, c, A, B, C are any real constants and X is a normal random variable with mean µ ∊ ℝ and standard deviation σ > 0, i.e., X ~ Norm(µ, σ2) with PDF ϕμ,σ(x)≡e−(x−μ)2/2σ2σ2π√,−∞<x<∞. The functions N(x) and N2(x, y; ρ) are the standard normal univariate and bivariate cumulative distribution functions, respectively.
E[eBXI{X>A}]≡∫∞AeBxϕμ,σ(x)dx=eμB+12σ2B2N(σB+μ−Aσ) (A.1)
E[eBXI{X<A}]≡∫A−∞eBxϕμ,σ(x)dx=eμB+12σ2B2N(−σB+A−μσ) (A.2)
E[N(AX+C)]≡∫∞−∞N(Ax+C)ϕμ,σ(x)dx=N(μA+C1+σ2A2√) (A.3)
E[eBXN(AX+C)]≡∫∞−∞eBxN(Ax+C)ϕμ,σ(x)dx=eμB+12σ2B2N(μA+C+σ2AB1+σ2A2√) (A.A)
∫∞0N(Ax+B)exdx=−N(B)+e(1−2AB)/2A2N(1−AB|A|)(forA<0) (A.5)
∫∞0N(Ax+B)e−xdx=N(B)+sgn (A)e(1+2AB)/2A2N(−1+AB|A|) (A.6)
∫∞−∞N(Ax+B)exdx=e(1−2AB)/2A2N(1−AB|A|)−e(1+2AB)/2A2N(−1+AB|A|)(forA<0) (A.7)
∫∞0N(Ax+B)dx=1|A|[BN(B)+e−B2/22π−−√](forA<0) (A.8)
E[N(AX+C)I{X<B}]≡∫B−∞N(Ax+C)ϕμ,σ(x)dx=N2(B−μσ,μA+C1+σ2A2√;−σA1+σ2A2√) (A.9)
E[N(AX+C)I{X>B}]≡∫∞BN(Ax+C)ϕμ,σ(x)dx=N2(μ−Bσ,μA+C1+σ2A2√;σA1+σ2A2√) (A.10)
Note that formulas (A.1)–(A.10) simplify in the special case when Xd¯¯Z~Norm(0,1) by setting µ = 0, σ = 1.
The bivariate normal CDF also has useful symmetry relations such as
N2(a,b;ρ)=N2(b,a;ρ), (A.11)
N2(a,b;ρ)+N2(a,−b;−ρ)=N(a). (A.12)
In the expectation formulas below, Z1,Z2 are i.i.d. standard normal random variables with covariance Cov(Z1,Z2)= ρ, |ρ| < 1, i.e., with joint PDF n2(x,y;ρ)≡∂2∂x∂yN2(x,y;ρ)=12π1−ρ2√e−(x2+y2−2ρxy)/2(1−ρ2).
E[e−BZ2I{Z1<a,Z2<b}]≡∫∫R2e−Byn2(x,y;ρ)I{x<a,y<b}dxdy=e12B2N2(a+ρB,b+B;ρ) (A.13)
E[e−BZ2I{Z1>a,Z2<b}]=e12B2N2(−(a+ρB),b+B;−ρ) (A.14)
E[e−BZ2I{Z1<a,Z2>b}]=e12B2N2(a+ρB,−(b+B);−ρ) (A.15)
E[e−BZ2I{Z1>a,Z2>b}]=e12B2N2(−(a+ρB),−(b+B);ρ) (A.16)
Note: interchanging a ↔ b in (A.13)–(A.16) gives respectively equivalent formulas for the expectations E[e−BZ1I{Z1<a,Z2<b}], E[e−BZ1I{Z1<a,Z2>b}], E[e−BZ1I{Z1>a,Z2<b}], and E[e−BZ1I{Z1>a,Z2>b}].