ISEN 310: Uncertainty Modeling for IE Department of Industrial and Systems Engineering Texas A&M University Fall 2023 Petar Momčilović petar@tamu.edu Expectation Expectation * I E[X] is a number: EX = I Examples (P Reiki x pX (x) 1 x fX (x) dx R 1x I Geometric X I Exponential X - I Geometry *N1 , I Random variable X if X discrete if X continuous bat 2 Expectation of g(X) g(X) Y #g(X) EY = I Random variable X = I Function g likelihood I Random g(X) E[g(X)] = value g(x) p (x) (P R 1x 1 g(x) fX (x) dx I Examples I E[X k ]: kth moment I k = 1: mean I E[(X X if X discrete if X continuous x EX)k ]: kth centered moment I k = 2: variance I E[esX ]: moment generating function (MGF) - Expectation = Integral /xx =() xxx)2 Saxdx I In general I Expectation is linear Eg(X) 6= g(EX) ((=x b) &x + d E[ag(X)] = aE[g(X)] E[g(X) + h(X)] = Eg(X) + Eh(X) I Variance I Motivation I Var(X) = E[X 2 ] (EX)2 I Example: Bernoulli r.v. I Properties Example buy x amountprofit--ex+p min2x D3 - E profit I Newsvendor model = , D1] EL-4x I Buy items for $c per unit I Sell items for $p per unit I Random demand D: CDF FD I Maximize expected profit I Optimal order quantity , qxD3 = Elg max E[g (D)] , Moment Generating Function - I & I MX (s) = E[esX ] I Moments Ms -M mi Is EX M's -- EE(X] I Examples I Geometric r.v. I Exponential r.v. I MX (s) = 41 e 3s + 12 e2s + 14 e5s ElesX] :. E - P(X 3) 4 5) 4(X 2) 1/2 4(X = - = = E es D(=x ) : * E(Xes ] *= " I MGF $ CDF -> 1E[-X es] = : = = = i Expectation of g(X, Y ) I Random variables X and Y (P m g(x, y) pX,Y (x, y) E[g(X, Y )] = R 1x 1 g(x, Y ) fX,Y (x, y) dx vahie Likelihood if (X, Y ) discrete if (X, Y ) continuous I Example: E[max{X, Y }] with X and Y random samples from {1, 2, . . . , 310} without replacement Covariance and Correlation Cor(X X) , = Var(X) I Random variables X and Y a number Cov(X, Y ) = E[(X EX)(Y EY )] -Cov(X, Y ) - -1 , 1] -> ⇢ = p Var(X) Var(Y ) I Uncorrelated 6= Independent I Example: Disk I Data I Expectation = Integral I Linearity E[g(X) + h(Y )] = E[g(X)] + E[h(Y )] en I Even if X and Y are not independent E F I Example: Hypergeometric distribution I n objects: k of them are faulty I Goal: estimate k I Sample m objects at random without replacement I X = number of faulty objects in the sample - Total Expectation I Conditional expectation (on an event) ~likelihood (PValue xp (x) E[X | A] = R 1x X|A 1 x fX|A (x) dx if X discrete if X continuous I Events B1 , B2 , . . .: Bi \ Bj = ; and [i Bi = ⌦ I Total Expectation Theorem: EX = X E[X | Bi ] P[Bi ] M I / B3 i - Examples EX / I Geometric r.v. I Branching process : 1 . definitio I MGF . 3 . - Conditional Expectation I Event {Y = y} - value d Likelihood (P Ne xp (x|y) E[X | Y = y] = h(y) = R 1x X|Y 2 1 x fX|Y (x|y) dx A if X discrete if X continuous / * rv I Random variable (!) / . E[X | Y ] = h(Y ) -> - - -K I Iterated expectation vondim variable tower - E[X] = E[E[X | Y ]] / a - property - - Examples I Stick I Insurance claims - Sum of Independent R.V.s I Recall: Independent r.v.s X and Y FX,Y (x, y) = FX (x) FY (y) I Implication E[g(X) h(Y )] = E[g(X)] E[h(Y )] en I Examples I Var(X + Y ) I Cov(X, Y ) MGF of a Sum I Independent r.v.s X and Y I MGF dif -get MX+Y (s) = E[es(X+Y ) ] = E[esX esY ] = E[esX ] E[esY ] - I Example: Normal r.v.s y My MGF of a Random Sum I Independent, identically distributed X1 , X2 , . . . I N : r.v. independent of X1 , X2 , . . . I Sum S= N X L Xi i=1 I MGF definiti ⇥ ⇤ e Eles] MS (s) = E esS ⇥ ⇤ = E E[esS | N ] ⇥ ⇤ = E (MX (s))N MNs) e I Algorithm I Find MX (s) I Replace es by MX (s) in MN (s) E([esN] = - Examples C S I Example 1 I Xi : exponential I N : geometric O I Example 2 I Xi : geometric I N : geometric - > = NX E O ~exp . : exercise