Probability and Statistics

Cookbook

Copyright
c Matthias Vallentin, 2015
vallentin@icir.org

31st March, 2015

This cookbook integrates a variety of topics in probability the- 12 Parametric Inference 13 20 Stochastic Processes 24
ory and statistics. It is based on literature and in-class material 12.1 Method of Moments . . . . . . . . . . . 13 20.1 Markov Chains . . . . . . . . . . . . . . 24
from courses of the statistics department at the University of 12.2 Maximum Likelihood . . . . . . . . . . . 14 20.2 Poisson Processes . . . . . . . . . . . . . 25
California in Berkeley but also influenced by other sources [2, 3]. 12.2.1 Delta Method . . . . . . . . . . . 14
If you find errors or have suggestions for further topics, I would 12.3 Multiparameter Models . . . . . . . . . 15 21 Time Series 25
appreciate if you send me an email. The most recent version 21.1 Stationary Time Series . . . . . . . . . . 26
12.3.1 Multiparameter delta method . . 15
of this document is available at http://matthias.vallentin. 21.2 Estimation of Correlation . . . . . . . . 26
12.4 Parametric Bootstrap . . . . . . . . . . 15
net/probability-and-statistics-cookbook/. To reproduce, 21.3 Non-Stationary Time Series . . . . . . . 26
please contact me. 21.3.1 Detrending . . . . . . . . . . . . 27
13 Hypothesis Testing 15
21.4 ARIMA models . . . . . . . . . . . . . . 27
14 Exponential Family 16 21.4.1 Causality and Invertibility . . . . 28
Contents 21.5 Spectral Analysis . . . . . . . . . . . . . 28
15 Bayesian Inference 16
1 Distribution Overview 3 22 Math 29
15.1 Credible Intervals . . . . . . . . . . . . . 16
1.1 Discrete Distributions . . . . . . . . . . 3 22.1 Gamma Function . . . . . . . . . . . . . 29
1.2 Continuous Distributions . . . . . . . . 5 15.2 Function of parameters . . . . . . . . . . 17
22.2 Beta Function . . . . . . . . . . . . . . . 29
15.3 Priors . . . . . . . . . . . . . . . . . . . 17 22.3 Series . . . . . . . . . . . . . . . . . . . 29
2 Probability Theory 8 15.3.1 Conjugate Priors . . . . . . . . . 17 22.4 Combinatorics . . . . . . . . . . . . . . 30
15.4 Bayesian Testing . . . . . . . . . . . . . 18
3 Random Variables 8
3.1 Transformations . . . . . . . . . . . . . 9 16 Sampling Methods 18
16.1 Inverse Transform Sampling . . . . . . . 18
4 Expectation 9 16.2 The Bootstrap . . . . . . . . . . . . . . 18
16.2.1 Bootstrap Confidence Intervals . 18
5 Variance 9
16.3 Rejection Sampling . . . . . . . . . . . . 19
6 Inequalities 10 16.4 Importance Sampling . . . . . . . . . . . 19

7 Distribution Relationships 10 17 Decision Theory 19

17.1 Risk . . . . . . . . . . . . . . . . . . . . 19
8 Probability and Moment Generating 17.2 Admissibility . . . . . . . . . . . . . . . 20
Functions 11 17.3 Bayes Rule . . . . . . . . . . . . . . . . 20
17.4 Minimax Rules . . . . . . . . . . . . . . 20
9 Multivariate Distributions 11
9.1 Standard Bivariate Normal . . . . . . . 11 18 Linear Regression 20
9.2 Bivariate Normal . . . . . . . . . . . . . 11 18.1 Simple Linear Regression . . . . . . . . 20
9.3 Multivariate Normal . . . . . . . . . . . 11 18.2 Prediction . . . . . . . . . . . . . . . . . 21
10 Convergence 11 18.3 Multiple Regression . . . . . . . . . . . 21
10.1 Law of Large Numbers (LLN) . . . . . . 12 18.4 Model Selection . . . . . . . . . . . . . . 22
10.2 Central Limit Theorem (CLT) . . . . . 12
19 Non-parametric Function Estimation 22
11 Statistical Inference 12 19.1 Density Estimation . . . . . . . . . . . . 22
11.1 Point Estimation . . . . . . . . . . . . . 12 19.1.1 Histograms . . . . . . . . . . . . 23
11.2 Normal-Based Confidence Interval . . . 13 19.1.2 Kernel Density Estimator (KDE) 23
11.3 Empirical distribution . . . . . . . . . . 13 19.2 Non-parametric Regression . . . . . . . 23
11.4 Statistical Functionals . . . . . . . . . . 13 19.3 Smoothing Using Orthogonal Functions 24
1 Distribution Overview
1.1 Discrete Distributions
Notation1 FX (x) fX (x) E [X] V [X] MX (s)

0 x<a
(b − a + 1)2 − 1 eas − e−(b+1)s

bxc−a+1 I(a ≤ x ≤ b) a+b
Uniform Unif {a, . . . , b} a≤x≤b
 b−a b−a+1 2 12 s(b − a)
1 x>b

Bernoulli Bern (p) (1 − p)1−x px (1 − p)1−x p p(1 − p) 1 − p + pes
!
n x
Binomial Bin (n, p) I1−p (n − x, x + 1) p (1 − p)n−x np np(1 − p) (1 − p + pes )n
x
k k
!n
n! x
X X
Multinomial Mult (n, p) px1 1 · · · pkk xi = n npi npi (1 − pi ) pi e si
x1 ! . . . xk ! i=1 i=0
! m m−x



x − np x n−x nm nm(N − n)(N − m)
Hypergeometric Hyp (N, m, n) ≈Φ N N 2 (N − 1)
p

np(1 − p) x
N
!  r
x+r−1 r 1−p 1−p p
Negative Binomial NBin (r, p) Ip (r, x + 1) p (1 − p)x r r
r−1 p p2 1 − (1 − p)es
1 1−p pes
Geometric Geo (p) 1 − (1 − p)x x ∈ N+ p(1 − p)x−1 x ∈ N+
p p2 1 − (1 − p)es
x
X λi λx e−λ s
Poisson Po (λ) e−λ λ λ eλ(e −1)

i=0
i! x!

1 We use the notation γ(s, x) and Γ(x) to refer to the Gamma functions (see §22.1), and use B(x, y) and Ix to refer to the Beta functions (see §22.2).

3
 Uniform (discrete) Binomial Geometric Poisson
●
● n = 40, p = 0.3 0.8 ●
● p = 0.2 ● ●
● λ=1
● n = 30, p = 0.6 ● p = 0.5 ● λ=4
● n = 25, p = 0.9 ● p = 0.8 ● λ = 10
●
0.3
0.2 ● 0.6

●
0.2
PMF

PMF

PMF

PMF
1 ● ● ● ●
● ●
● ● ● ● ● ● ● 0.4 ●
n ●
● ● ●
●
● ●
0.1
● ●
● ● ● ● ●
● ●
● 0.1 ●
● 0.2 ● ●
● ● ● ● ●
● ●
● ● ●
● ●
● ● ●
● ● ● ● ● ●
● ● ●
● ● ● ● ● ●
● ● ● ● ● ● ● ●
● ●●● ● ● ● ●
0.0 ●●●● ●●
●●●●●●● ●●●●●●● ●●●●●●●●●●●●●●●●●●●●● 0.0 ● ●
●
● ●
● ● ● ● ● 0.0 ● ● ● ● ● ● ● ● ● ●
● ● ● ● ● ● ● ● ● ●
● ●

a b 0 10 20 30 40 0.0 2.5 5.0 7.5 10.0 0 5 10 15 20

x x x x
Uniform (discrete) Binomial Geometric Poisson
1 ● 1.00 ●●●●●●●●●●●●●●●●
● ● ●●●●●●●●●●●●●●●●●●●●●
●● 1.0 ● ● ● ● ●
● ● ● ● ● 1.00 ● ● ● ● ● ●
● ●
● ● ● ● ● ● ● ● ●
● ● ●
● ●
● ● ● ●
● ● ● ●
● ● ● ● ●
● ●
● ●
● ●
● ●
●
●
● ● ●
0.75 ●
0.8 ● ● 0.75 ●
● ● ●
i ●
●
● ●
n ●
● ● ●
●
CDF

CDF

CDF

CDF
0.50 0.6 ● 0.50
● ●
● ●
●
● ●
i ●
● ● ●
n ●
●
0.25 ● 0.4 0.25 ●
●
● ●
●
●
● ● ● ● n = 40, p = 0.3 ● p = 0.2 ● ● λ=1
● ● ●
● n = 30, p = 0.6 ● p = 0.5 ●
● λ=4
●
0 ● 0.00 ●●●● ●
●
●●
●●●●●●●●●●●●●●●●●
●●●●●●●●●● ● ● n = 25, p = 0.9 0.2 ● ● p = 0.8 0.00
●
● ● ● ● ● λ = 10

a b 0 10 20 30 40 0.0 2.5 5.0 7.5 10.0 0 5 10 15 20

x x x x

4
1.2 Continuous Distributions
Notation FX (x) fX (x) E [X] V [X] MX (s)

0 x<a
(b − a)2 esb − esa

x−a I(a < x < b) a+b
Uniform Unif (a, b) a<x<b
 b−a b−a 2 12 s(b − a)
1 x>b

(x − µ)2
Z x
σ 2 s2
   
1
N µ, σ 2 σ2


Normal Φ(x) = φ(t) dt φ(x) = √ exp − µ exp µs +
−∞ σ 2π 2σ 2 2
(ln x − µ)2
   
1 1 ln x − µ 1 2 2 2
ln N µ, σ 2 eµ+σ /2
(eσ − 1)e2µ+σ


Log-Normal + erf √ √ exp −
2 2 2σ 2 x 2πσ 2 2σ 2
 
1 T
Σ−1 (x−µ) 1
Multivariate Normal MVN (µ, Σ) (2π)−k/2 |Σ|−1/2 e− 2 (x−µ) µ Σ exp µT s + sT Σs
2
−(ν+1)/2 ( ν
Γ ν+1


ν ν 
2 x2 ν−2
ν>2
Student’s t Student(ν) Ix , √ 1 + 0
νπΓ ν2


2 2 ν ∞ 1<ν≤2
 
1 k x 1
Chi-square χ2k γ , xk/2−1 e−x/2 k 2k (1 − 2s)−k/2 s < 1/2
Γ(k/2) 2 2 2k/2 Γ(k/2)
r
d
(d1 x)d1 d2 2
2d22 (d1 + d2 − 2)
 
d1 d1 (d1 x+d2 )d1 +d2 d2
F F(d1 , d2 ) I d1 x , d1 d1 d2 − 2 d1 (d2 − 2)2 (d2 − 4)


d1 x+d2 2 2 xB 2
, 2
1 −x/β 1
Exponential Exp (β) 1 − e−x/β e β β2 (s < 1/β)
β 1 − βs
 α
γ(α, x/β) 1 1
Gamma Gamma (α, β) xα−1 e−x/β αβ αβ 2 (s < 1/β)
Γ(α) Γ (α) β α 1 − βs
Γ α, βx


β α −α−1 −β/x β β2 2(−βs)α/2 p 
Inverse Gamma InvGamma (α, β) x e α>1 α>2 Kα −4βs
Γ (α) Γ (α) α−1 (α − 1)2 (α − 2) Γ(α)
P 
k
Γ i=1 αi Y α −1
k
αi E [Xi ] (1 − E [Xi ])
Dirichlet Dir (α) Qk xi i Pk Pk
i=1 Γ (αi ) i=1 i=1 αi i=1 αi + 1
∞ k−1
!
Γ (α + β) α−1 α αβ X Y α+r sk
Beta Beta (α, β) Ix (α, β) x (1 − x)β−1 1+
Γ (α) Γ (β) α+β (α + β)2 (α + β + 1) r=0
α+β+r k!
k=1
    ∞ n n
k k  x k−1 −(x/λ)k 1 2 X s λ  n
Weibull Weibull(λ, k) 1 − e−(x/λ) e λΓ 1 + λ2 Γ 1 + − µ2 Γ 1+
λ λ k k n=0
n! k
 x α
m xα αxm xα
Pareto Pareto(xm , α) 1− x ≥ xm m
α α+1 x ≥ xm α>1 m
α>2 α(−xm s)α Γ(−α, −xm s) s < 0
x x α−1 (α − 1)2 (α − 2)

5
 Uniform (continuous) Normal Log−Normal Student's t
2.0 1.00 0.4
µ = 0, σ2 = 0.2 µ = 0, σ2 = 3 ν=1
µ = 0, σ2 = 1 µ = 2, σ2 = 2 ν=2
µ = 0, σ2 = 5 µ = 0, σ2 = 1 ν=5
ν=∞
µ = −2, σ2 = 0.5 µ = 0.5, σ2 = 1
µ = 0.25, σ2 = 1
1.5 0.75 µ = 0.125, σ2 = 1 0.3
PDF

PDF

PDF

PDF
1
● ● 1.0 0.50 0.2
b−a

0.5 0.25 0.1

● ● 0.0 0.00 0.0

a b −5.0 −2.5 0.0 2.5 5.0 0 1 2 3 −5.0 −2.5 0.0 2.5 5.0
x x x x
χ 2 F Exponential Gamma
d1 = 1, d2 = 1 2.0 β=2 2.0 α = 1, β = 2
1.00 k=1 3 d1 = 2, d2 = 1 β=1 α = 2, β = 2
k=2 d1 = 5, d2 = 2 β = 0.4 α = 3, β = 2
k=3 d1 = 100, d2 = 1 α = 5, β = 1
k=4 d1 = 100, d2 = 100 α = 9, β = 0.5
k=5
1.5 1.5
0.75

2
PDF

PDF

PDF
PDF

0.50 1.0 1.0

1
0.25 0.5 0.5

0.00 0 0.0 0.0

0 2 4 6 8 0 1 2 3 4 5 0 1 2 3 4 5 0 5 10 15 20
x x x x
Inverse Gamma Beta Weibull Pareto
α = 1, β = 1 5 α = 0.5, β = 0.5 2.0 λ = 1, k = 0.5 4 xm = 1, k = 1
α = 2, β = 1 α = 5, β = 1 λ = 1, k = 1 xm = 1, k = 2
α = 3, β = 1 α = 1, β = 3 λ = 1, k = 1.5 xm = 1, k = 4
4 α = 3, β = 0.5 α = 2, β = 2 λ = 1, k = 5
4 α = 2, β = 5
1.5 3

3
3
PDF

PDF

PDF

PDF
1.0 2
2
2

0.5 1
1 1

0 0 0.0 0

0 1 2 3 4 5 0.00 0.25 0.50 0.75 1.00 0.0 0.5 1.0 1.5 2.0 2.5 1.0 1.5 2.0 2.5
x x x x
6
 Uniform (continuous) Normal Log−Normal Student's t
1 1.00 1.00
µ = 0, σ2 = 3
µ = 2, σ2 = 2
0.75 µ = 0, σ2 = 1
µ = 0.5, σ2 = 1
µ = 0.25, σ2 = 1
0.75 µ = 0.125, σ2 = 1 0.75

0.50
CDF

CDF

CDF

CDF
0.50 0.50

0.25
0.25 0.25

µ = 0, σ = 0.2
2
ν=1
µ = 0, σ2 = 1 ν=2
µ = 0, σ2 = 5 ν=5
0 0.00 µ = −2, σ2 = 0.5 0.00 0.00 ν=∞

a b −5.0 −2.5 0.0 2.5 5.0 0 1 2 3 −5.0 −2.5 0.0 2.5 5.0
x x x x
χ 2 F Exponential Gamma
1.00 1.00 1.00
1.00

0.75 0.75 0.75 0.75

CDF

CDF

CDF
CDF

0.50 0.50 0.50 0.50

0.25 0.25 0.25 0.25

k=1 d1 = 1, d2 = 1 α = 1, β = 2
k=2 d1 = 2, d2 = 1 α = 2, β = 2
k=3 d1 = 5, d2 = 2 β=2 α = 3, β = 2
k=4 d1 = 100, d2 = 1 β=1 α = 5, β = 1
0.00 k=5 0.00 d1 = 100, d2 = 100 0.00 β = 0.4 0.00 α = 9, β = 0.5

0 2 4 6 8 0 1 2 3 4 5 0 1 2 3 4 5 0 5 10 15 20
x x x x
Inverse Gamma Beta Weibull Pareto
1.00 1.00
1.00 1.00 α = 0.5, β = 0.5
α = 5, β = 1
α = 1, β = 3
α = 2, β = 2
α = 2, β = 5
0.75 0.75 0.75 0.75
CDF

CDF

CDF

CDF
0.50 0.50 0.50 0.50

0.25 0.25 0.25 0.25

α = 1, β = 1 λ = 1, k = 0.5
α = 2, β = 1 λ = 1, k = 1 xm = 1, k = 1
α = 3, β = 1 λ = 1, k = 1.5 xm = 1, k = 2
0.00 α = 3, β = 0.5 0.00 0.00 λ = 1, k = 5 0.00 xm = 1, k = 4

0 1 2 3 4 5 0.00 0.25 0.50 0.75 1.00 0.0 0.5 1.0 1.5 2.0 2.5 1.0 1.5 2.0 2.5
x x x x
7
2 Probability Theory Law of Total Probability
n n
Definitions X G
P [B] = P [B|Ai ] P [Ai ] Ω= Ai
• Sample space Ω i=1 i=1

• Outcome (point or element) ω ∈ Ω Bayes’ Theorem

• Event A ⊆ Ω n
• σ-algebra A P [B | Ai ] P [Ai ] G
P [Ai | B] = Pn Ω= Ai
1. ∅ ∈ A j=1 P [B | Aj ] P [Aj ] i=1
S∞
2. A1 , A2 , . . . , ∈ A =⇒ i=1 Ai ∈ A Inclusion-Exclusion Principle
3. A ∈ A =⇒ ¬A ∈ A
 n n
 r
[  X X \ 
• Probability Distribution P (−1)r−1
 
 Ai  =  A ij 

  
1. P [A] ≥ 0 ∀A i=1 r=1 i≤i1 <···<ir ≤n j=1

2. P [Ω] = 1
"∞ #
G ∞
X 3 Random Variables
3. P Ai = P [Ai ]
i=1 i=1 Random Variable (RV)
• Probability space (Ω, A, P) X:Ω→R
Probability Mass Function (PMF)
Properties
• P [∅] = 0 fX (x) = P [X = x] = P [{ω ∈ Ω : X(ω) = x}]
• B = Ω ∩ B = (A ∪ ¬A) ∩ B = (A ∩ B) ∪ (¬A ∩ B) Probability Density Function (PDF)
• P [¬A] = 1 − P [A]
Z b
• P [B] = P [A ∩ B] + P [¬A ∩ B]
P [a ≤ X ≤ b] = f (x) dx
• P [Ω] = 1 P [∅] = 0 a
S T T S
• ¬( n An ) = n ¬An ¬( n An ) = n ¬An DeMorgan
S T Cumulative Distribution Function (CDF)
• P [ n An ] = 1 − P [ n ¬An ]
• P [A ∪ B] = P [A] + P [B] − P [A ∩ B] FX : R → [0, 1] FX (x) = P [X ≤ x]
=⇒ P [A ∪ B] ≤ P [A] + P [B] 1. Nondecreasing: x1 < x2 =⇒ F (x1 ) ≤ F (x2 )
• P [A ∪ B] = P [A ∩ ¬B] + P [¬A ∩ B] + P [A ∩ B] 2. Normalized: limx→−∞ = 0 and limx→∞ = 1
• P [A ∩ ¬B] = P [A] − P [A ∩ B] 3. Right-Continuous: limy↓x F (y) = F (x)
Continuity of Probabilities
S∞ Z b
• A1 ⊂ A2 ⊂ . . . =⇒ limn→∞ P [An ] = P [A] whereA = i=1 Ai P [a ≤ Y ≤ b | X = x] = fY |X (y | x)dy a≤b
T∞
• A1 ⊃ A2 ⊃ . . . =⇒ limn→∞ P [An ] = P [A] whereA = i=1 Ai a

f (x, y)
Independence ⊥
⊥ fY |X (y | x) =
A⊥
⊥ B ⇐⇒ P [A ∩ B] = P [A] P [B] fX (x)
Independence
Conditional Probability
1. P [X ≤ x, Y ≤ y] = P [X ≤ x] P [Y ≤ y]
P [A ∩ B]
P [A | B] = P [B] > 0 2. fX,Y (x, y) = fX (x)fY (y)
P [B] 8
 Z
3.1 Transformations • E [XY ] = xyfX,Y (x, y) dFX (x) dFY (y)
X,Y
Transformation function
• E [ϕ(Y )] 6= ϕ(E [X]) (cf. Jensen inequality)
Z = ϕ(X)
• P [X ≥ Y ] = 1 =⇒ E [X] ≥ E [Y ]
Discrete • P [X = Y ] = 1 ⇐⇒ E [X] = E [Y ]
X ∞
fZ (z) = P [ϕ(X) = z] = P [{x : ϕ(x) = z}] = P X ∈ ϕ−1 (z) =
 
f (x)
X
• E [X] = P [X ≥ x]
x∈ϕ−1 (z) x=1

Continuous Sample mean

n
Z 1X
X̄n = Xi
FZ (z) = P [ϕ(X) ≤ z] = f (x) dx with Az = {x : ϕ(x) ≤ z} n i=1
Az
Conditional expectation
Special case if ϕ strictly monotone Z

d
  
 dx  1 • E [Y | X = x] = yf (y | x) dy
fZ (z) = fX (ϕ−1 (z))  ϕ−1 (z) = fX (x)   = fX (x)

dz dz |J| • E [X] = E [E [X | Y ]]
Z ∞
The Rule of the Lazy Statistician • E[ϕ(X, Y ) | X = x] = ϕ(x, y)fY |X (y | x) dx
Z Z −∞
∞
E [Z] = ϕ(x) dFX (x) • E [ϕ(Y, Z) | X = x] = ϕ(y, z)f(Y,Z)|X (y, z | x) dy dz
−∞
Z Z • E [Y + Z | X] = E [Y | X] + E [Z | X]
E [IA (x)] = IA (x) dFX (x) = dFX (x) = P [X ∈ A] • E [ϕ(X)Y | X] = ϕ(X)E [Y | X]
A
• E[Y | X] = c =⇒ Cov [X, Y ] = 0
Convolution
Z ∞ Z z
X,Y ≥0
• Z := X + Y fZ (z) = fX,Y (x, z − x) dx = fX,Y (x, z − x) dx
−∞ 0 5 Variance
Z ∞
• Z := |X − Y | fZ (z) = 2 fX,Y (x, z + x) dx Definition and properties
0
Z ∞ Z ∞ 2
    2
X ⊥⊥ • V [X] = σX = E (X − E [X])2 = E X 2 − E [X]
• Z := fZ (z) = |x|fX,Y (x, xz) dx = xfx (x)fX (x)fY (xz) dx " n # n
Y −∞ −∞ X X X
• V Xi = V [Xi ] + 2 Cov [Xi , Yj ]
i=1 i=1 i6=j
4 Expectation " n
X
# n
X
• V Xi = V [Xi ] if Xi ⊥
⊥ Xj
Definition and properties i=1 i=1
X

 xfX (x) X discrete Standard deviation p
sd[X] = V [X] = σX

Z  x

• E [X] = µX = x dFX (x) = Covariance

 Z
 xfX (x) dx X continuous


• Cov [X, Y ] = E [(X − E [X])(Y − E [Y ])] = E [XY ] − E [X] E [Y ]
• P [X = c] = 1 =⇒ E [X] = c • Cov [X, a] = 0
• E [cX] = c E [X] • Cov [X, X] = V [X]
• E [X + Y ] = E [X] + E [Y ] • Cov [X, Y ] = Cov [Y, X]
9
 • Cov [aX, bY ] = abCov [X, Y ] 7 Distribution Relationships
• Cov [X + a, Y + b] = Cov [X, Y ]

n m

n X m
Binomial
X X X
n
• Cov  Xi , Yj  = Cov [Xi , Yj ] X
i=1 j=1 i=1 j=1
• Xi ∼ Bern (p) =⇒ Xi ∼ Bin (n, p)
i=1
Correlation • X ∼ Bin (n, p) , Y ∼ Bin (m, p) =⇒ X + Y ∼ Bin (n + m, p)
Cov [X, Y ]
ρ [X, Y ] = p • limn→∞ Bin (n, p) = Po (np) (n large, p small)
V [X] V [Y ] • limn→∞ Bin (n, p) = N (np, np(1 − p)) (n large, p far from 0 and 1)
Independence
Negative Binomial
X⊥
⊥ Y =⇒ ρ [X, Y ] = 0 ⇐⇒ Cov [X, Y ] = 0 ⇐⇒ E [XY ] = E [X] E [Y ]
• X ∼ NBin (1, p) = Geo (p)
Pr
Sample variance • X ∼ NBin (r, p) = i=1 Geo (p)
n P P
1 X • Xi ∼ NBin (ri , p) =⇒ Xi ∼ NBin ( ri , p)
S2 = (Xi − X̄n )2
n − 1 i=1 • X ∼ NBin (r, p) . Y ∼ Bin (s + r, p) =⇒ P [X ≤ s] = P [Y ≥ r]

Conditional variance Poisson

n n
!
    2
• V [Y | X] = E (Y − E [Y | X])2 | X = E Y 2 | X − E [Y | X] X X
• Xi ∼ Po (λi ) ∧ Xi ⊥⊥ Xj =⇒ Xi ∼ Po λi
• V [Y ] = E [V [Y | X]] + V [E [Y | X]] i=1 i=1
  
 n n
X X λ i
6 Inequalities • Xi ∼ Po (λi ) ∧ Xi ⊥⊥ Xj =⇒ Xi  Xj ∼ Bin  Xj , Pn 
 j=1 j=1 j=1 λ j

Cauchy-Schwarz
2 Exponential
E [XY ] ≤ E X 2 E Y 2
   
n
X
Markov • Xi ∼ Exp (β) ∧ Xi ⊥
⊥ Xj =⇒ Xi ∼ Gamma (n, β)
E [ϕ(X)]
P [ϕ(X) ≥ t] ≤ i=1
t • Memoryless property: P [X > x + y | X > y] = P [X > x]
Chebyshev
V [X] Normal
P [|X − E [X]| ≥ t] ≤
t2

X−µ

Chernoff • X ∼ N µ, σ 2 =⇒ σ∼ N (0, 1)
δ
 
e •


X ∼ N µ, σ 2 ∧ Z = aX + b =⇒ Z ∼ N aµ + b, a2 σ 2


P [X ≥ (1 + δ)µ] ≤ δ > −1
(1 + δ)1+δ •



X ∼ N µ1 , σ12 ∧ Y ∼ N µ2 , σ22 =⇒ X + Y ∼ N µ1 + µ2 , σ12 + σ22



P 2

Xi ∼ N µi , σi2 =⇒
Hoeffding P P
• X ∼N i µi , i σi
 i i
b−µ a−µ


X1 , . . . , Xn independent ∧ P [Xi ∈ [ai , bi ]] = 1 ∧ 1 ≤ i ≤ n • P [a < X ≤ b] = Φ σ − Φ σ
2 • Φ(−x) = 1 − Φ(x) φ0 (x) = −xφ(x) φ00 (x) = (x2 − 1)φ(x)
P X̄ − E X̄ ≥ t ≤ e−2nt t > 0
   
• Upper quantile of N (0, 1): zα = Φ−1 (1 − α)
2n2 t2
 
   
P |X̄ − E X̄ | ≥ t ≤ 2 exp − Pn 2
t>0 Gamma
i=1 (bi − ai )
Jensen • X ∼ Gamma (α, β) ⇐⇒ X/β ∼ Gamma (α, 1)
Pα
E [ϕ(X)] ≥ ϕ(E [X]) ϕ convex • Gamma (α, β) ∼ i=1 Exp (β)
10
 9.2 Bivariate Normal
P P
• Xi ∼ Gamma (αi , β) ∧ Xi ⊥
⊥ Xj =⇒ i Xi ∼ Gamma ( i αi , β)
Z ∞
Γ(α)


• = xα−1 e−λx dx Let X ∼ N µx , σx2 and Y ∼ N µy , σy2 .
λα 0
 
Beta 1 z
f (x, y) = exp −
2(1 − ρ2 )
p
2πσx σy 1−ρ 2
1 Γ(α + β) α−1
• xα−1 (1 − x)β−1 = x (1 − x)β−1
B(α, β) Γ(α)Γ(β) " 2  2   #
  B(α + k, β) α+k−1 x − µx y − µy x − µx y − µy
• E Xk = = E X k−1
  z= + − 2ρ
B(α, β) α+β+k−1 σx σy σx σy
• Beta (1, 1) ∼ Unif (0, 1) Conditional mean and variance
σX
E [X | Y ] = E [X] + ρ (Y − E [Y ])
8 Probability and Moment Generating Functions σY
p
V [X | Y ] = σX 1 − ρ2
 
• GX (t) = E tX |t| < 1
"∞ # ∞  
X (Xt)i X E Xi
· ti
t
 Xt

• MX (t) = GX (e ) = E e =E = 9.3 Multivariate Normal
i=0
i! i=0
i!
• P [X = 0] = GX (0) Covariance matrix Σ (Precision matrix Σ−1 )
• P [X = 1] = G0X (0)  
(i)
V [X1 ] · · · Cov [X1 , Xk ]
GX (0) .. .. ..
• P [X = i] = Σ=
 
. . . 
i!
• E [X] = G0X (1− ) Cov [Xk , X1 ] · · · V [Xk ]
  (k)
• E X k = MX (0) If X ∼ N (µ, Σ),
 
X! (k)
• E = GX (1− ) −1/2

1

(X − k)! fX (x) = (2π)−n/2 |Σ| exp − (x − µ)T Σ−1 (x − µ)
2 2
• V [X] = G00X (1− ) + G0X (1− ) − (G0X (1− ))
d Properties
• GX (t) = GY (t) =⇒ X = Y
• Z ∼ N (0, 1) ∧ X = µ + Σ1/2 Z =⇒ X ∼ N (µ, Σ)
9 Multivariate Distributions • X ∼ N (µ, Σ) =⇒ Σ−1/2 (X − µ) ∼ N (0, 1)


• X ∼ N (µ, Σ) =⇒ AX ∼ N Aµ, AΣAT
9.1 Standard Bivariate Normal


• X ∼ N (µ, Σ) ∧ kak = k =⇒ aT X ∼ N aT µ, aT Σa
p
Let X, Y ∼ N (0, 1) ∧ X ⊥
⊥ Z where Y = ρX + 1 − ρ2 Z
10 Convergence
Joint density
x2 + y 2 − 2ρxy
 
1 Let {X1 , X2 , . . .} be a sequence of rv’s and let X be another rv. Let Fn denote
f (x, y) = exp −
2(1 − ρ2 )
p
2π 1 − ρ2 the cdf of Xn and let F denote the cdf of X.
Conditionals
Types of convergence
(Y | X = x) ∼ N ρx, 1 − ρ2 (X | Y = y) ∼ N ρy, 1 − ρ2



and D
1. In distribution (weakly, in law): Xn → X
Independence
X⊥
⊥ Y ⇐⇒ ρ = 0 lim Fn (t) = F (t) ∀t where F continuous
n→∞ 11
 P
2. In probability: Xn → X 10.2 Central Limit Theorem (CLT)
Let {X1 , . . . , Xn } be a sequence of iid rv’s, E [X1 ] = µ, and V [X1 ] = σ 2 .
(∀ε > 0) lim P [|Xn − X| > ε] = 0
n→∞

√
3. Almost surely (strongly): Xn → X
as
X̄n − µ n(X̄n − µ) D
Zn := q   = →Z where Z ∼ N (0, 1)
V X̄n σ
h i h i
P lim Xn = X = P ω ∈ Ω : lim Xn (ω) = X(ω) = 1
n→∞ n→∞ lim P [Zn ≤ z] = Φ(z) z∈R
n→∞
qm
4. In quadratic mean (L2 ): Xn → X CLT notations

lim E (Xn − X)2 = 0 Zn ≈ N (0, 1)

 
n→∞
σ2
 
X̄n ≈ N µ,
Relationships n
σ2
 
qm P D X̄n − µ ≈ N 0,
• Xn → X =⇒ Xn → X =⇒ Xn → X n
as
• Xn → X =⇒ Xn → X
P √ 2


n(X̄n − µ) ≈ N 0, σ
D P
• Xn → X ∧ (∃c ∈ R) P [X = c] = 1 =⇒ Xn → X √
n(X̄n − µ)
• Xn
P
→X ∧ Yn
P
→ Y =⇒ Xn + Yn → X + Y
P
≈ N (0, 1)
qm qm qm
σ
• Xn →X ∧ Yn → Y =⇒ Xn + Yn → X + Y
P P P
• Xn →X ∧ Yn → Y =⇒ Xn Yn → XY
P P
• Xn →X =⇒ ϕ(Xn ) → ϕ(X) Continuity correction
D D
• Xn → X =⇒ ϕ(Xn ) → ϕ(X)  

x + 12 − µ

qm
• Xn → b ⇐⇒ limn→∞ E [Xn ] = b ∧ limn→∞ V [Xn ] = 0 P X̄n ≤ x ≈ Φ √
qm
σ/ n
• X1 , . . . , Xn iid ∧ E [X] = µ ∧ V [X] < ∞ ⇐⇒ X̄n → µ
x − 12 − µ
 
 
Slutzky’s Theorem P X̄n ≥ x ≈ 1 − Φ √
σ/ n
D P
• Xn → X and Yn → c =⇒ Xn + Yn → X + c
D
Delta method
D P D
• Xn → X and Yn → c =⇒ Xn Yn → cX 
σ2
 
2 σ2

0
D D
• In general: Xn → X and Yn → Y =⇒
6
D
Xn + Yn → X + Y Yn ≈ N µ, =⇒ ϕ(Yn ) ≈ N ϕ(µ), (ϕ (µ))
n n

10.1 Law of Large Numbers (LLN) 11 Statistical Inference

iid
Let {X1 , . . . , Xn } be a sequence of iid rv’s, E [X1 ] = µ. Let X1 , · · · , Xn ∼ F if not otherwise noted.

Weak (WLLN) 11.1 Point Estimation

P
X̄n → µ n→∞ • Point estimator θbn of θ is a rv: θbn = g(X1 , . . . , Xn )
h i
Strong (SLLN) • bias(θbn ) = E θbn − θ
as P
X̄n → µ n→∞ • Consistency: θbn → θ
12
 • Sampling distribution: F (θbn ) Nonparametric 1 − α confidence band for F
r h i
• Standard error: se(θbn ) = V θbn L(x) = max{Fbn − n , 0}
h i h i
• Mean squared error: mse = E (θbn − θ)2 = bias(θbn )2 + V θbn U (x) = min{Fbn + n , 1}
s  
• limn→∞ bias(θbn ) = 0 ∧ limn→∞ se(θbn ) = 0 =⇒ θbn is consistent 1 2
= log
θbn − θ D 2n α
• Asymptotic normality: → N (0, 1)
se
• Slutzky’s Theorem often lets us replace se(θbn ) by some (weakly) consis-
tent estimator σ
bn . P [L(x) ≤ F (x) ≤ U (x) ∀x] ≥ 1 − α

11.2 Normal-Based Confidence Interval 11.4 Statistical Functionals

  • Statistical functional: T (F )
b 2 . Let zα/2 = Φ−1 (1 − (α/2)), i.e., P Z > zα/2 = α/2
 
Suppose θbn ≈ N θ, se
  • Plug-in estimator of θ = (F ): θbn = T (Fbn )
and P −zα/2 < Z < zα/2 = 1 − α where Z ∼ N (0, 1). Then R
• Linear functional: T (F ) = ϕ(x) dFX (x)
• Plug-in estimator for linear functional:
Cn = θbn ± zα/2 se
b
Z n
1X
T (Fbn ) = ϕ(x) dFbn (x) = ϕ(Xi )
n i=1
11.3 Empirical distribution
 
Empirical Distribution Function (ECDF) b 2 =⇒ T (Fbn ) ± zα/2 se
• Often: T (Fbn ) ≈ N T (F ), se b
Pn • pth quantile: F −1 (p) = inf{x : F (x) ≥ p}
i=1 I(Xi ≤ x)
Fbn (x) = • µb = X̄n
n n
1 X
b2 =
• σ (Xi − X̄n )2
( n − 1 i=1
1 Xi ≤ x 1
Pn
I(Xi ≤ x) = n i=1 (Xi − µb)3
0 Xi > x • κ
b=
b3
Pσ n
i=1 (Xi − X̄n )(Yi − Ȳn )
Properties (for any fixed x) • ρb = qP qP
n 2 n 2
h i i=1 (Xi − X̄n ) i=1 (Yi − Ȳn )
• E Fbn = F (x)
h i F (x)(1 − F (x))
• V Fbn = 12 Parametric Inference
n
F (x)(1 − F (x)) D

Let F = f (x; θ) : θ ∈ Θ be a parametric model with parameter space Θ ⊂ Rk
• mse = →0
n and parameter θ = (θ1 , . . . , θk ).
P
• Fbn → F (x)
12.1 Method of Moments
Dvoretzky-Kiefer-Wolfowitz (DKW) inequality (X1 , . . . , Xn ∼ F )
  j th moment Z
  2
P sup F (x) − Fbn (x) > ε = 2e−2nε αj (θ) = E X j = xj dFX (x)
   
x 13
j th sample moment Fisher information (exponential family)
n
1X j
α
bj = X 
∂

n i=1 i I(θ) = Eθ − s(X; θ)
∂θ
Method of moments estimator (MoM)
Observed Fisher information
α1 (θ) = α
b1
n
α2 (θ) = α ∂2 X
Inobs (θ) = −
b2
log f (Xi ; θ)
.. .. ∂θ2 i=1
.=.
αk (θ) = α
bk Properties of the mle
Properties of the MoM estimator P
• Consistency: θbn → θ
• θbn exists with probability tending to 1 • Equivariance: θbn is the mle =⇒ ϕ(θbn ) ist the mle of ϕ(θ)
•
P
Consistency: θbn → θ • Asymptotic normality:
• Asymptotic normality:
p
1. se ≈ 1/In (θ)
√ D (θbn − θ) D
n(θb − θ) → N (0, Σ) → N (0, 1)
  se
where Σ = gE Y Y T g T , Y = (X, X 2 , . . . , X k )T , q
∂ −1 b ≈ 1/In (θbn )
2. se
g = (g1 , . . . , gk ) and gj = ∂θ αj (θ)
(θbn − θ) D
→ N (0, 1)
12.2 Maximum Likelihood se
b
Likelihood: Ln : Θ → [0, ∞) • Asymptotic optimality (or efficiency), i.e., smallest variance for large sam-
ples. If θen is any other estimator, the asymptotic relative efficiency is
n
Y
Ln (θ) = f (Xi ; θ) h i
i=1 V θbn
are(θen , θbn ) = h i ≤ 1
V θen
Log-likelihood
n
X
`n (θ) = log Ln (θ) = log f (Xi ; θ) • Approximately the Bayes estimator
i=1

Maximum likelihood estimator (mle)

12.2.1 Delta Method
Ln (θbn ) = sup Ln (θ) b where ϕ is differentiable and ϕ0 (θ) 6= 0:
If τ = ϕ(θ)
θ

Score function τn − τ ) D
(b
∂ → N (0, 1)
s(X; θ) = log f (X; θ) se(b
b τ)
∂θ
Fisher information where τb = ϕ(θ)
b is the mle of τ and
I(θ) = Vθ [s(X; θ)]
 
In (θ) = nI(θ) b = ϕ0 (θ)
se  se(
b θn )
 b b
14
12.3 Multiparameter Models 13 Hypothesis Testing
Let θ = (θ1 , . . . , θk ) and θb = (θb1 , . . . , θbk ) be the mle.
H0 : θ ∈ Θ0 versus H1 : θ ∈ Θ1
∂ 2 `n ∂ 2 `n
Hjj = Hjk = Definitions
∂θ2 ∂θj ∂θk
Fisher information matrix • Null hypothesis H0
• Alternative hypothesis H1
 
Eθ [H11 ] · · · Eθ [H1k ]
In (θ) = − 
 .. .. ..  • Simple hypothesis θ = θ0
. . . 
• Composite hypothesis θ > θ0 or θ < θ0
Eθ [Hk1 ] · · · Eθ [Hkk ]
• Two-sided test: H0 : θ = θ0 versus H1 : θ 6= θ0
Under appropriate regularity conditions • One-sided test: H0 : θ ≤ θ0 versus H1 : θ > θ0
(θb − θ) ≈ N (0, Jn ) • Critical value c
• Test statistic T
with Jn (θ) = In−1 . Further, if θbj is the j th component of θ, then • Rejection region R = {x : T (x) > c}
• Power function β(θ) = P [X ∈ R]
(θbj − θj ) D
→ N (0, 1) • Power of a test: 1 − P [Type II error] = 1 − β = inf β(θ)
se
bj θ∈Θ1
h i • Test size: α = P [Type I error] = sup β(θ)
b 2j = Jn (j, j) and Cov θbj , θbk = Jn (j, k)
where se θ∈Θ0

Retain H0 Reject H0
12.3.1 Multiparameter delta method √
H0 true Type
√ I Error (α)
Let τ = ϕ(θ1 , . . . , θk ) and let the gradient of ϕ be H1 true Type II Error (β) (power)
p-value
∂ϕ
 

 ∂θ1  • p-value = supθ∈Θ0 Pθ [T (X) ≥ T (x)] = inf α : T (x) ∈ Rα
 . 
∇ϕ =  ..  Pθ [T (X ? ) ≥ T (X)]

• p-value = supθ∈Θ0 = inf α : T (X) ∈ Rα

 ∂ϕ  | {z }
1−Fθ (T (X)) since T (X ? )∼Fθ
∂θk
 p-value evidence
Suppose ∇ϕθ=θb 6= 0 and τb = ϕ(θ).
b Then,
< 0.01 very strong evidence against H0
τ − τ) D
(b 0.01 − 0.05 strong evidence against H0
→ N (0, 1) 0.05 − 0.1 weak evidence against H0
se(b
b τ)
> 0.1 little or no evidence against H0
where r Wald test
 T  
se(b
b τ) = ∇ϕ
b Jbn ∇ϕ
b
• Two-sided test
θb − θ0

b and ∇ϕ
and Jbn = Jn (θ) b = ∇ϕ b.
θ=θ • Reject H0 when |W | > zα/2 where W =
  se
b
12.4 Parametric Bootstrap • P |W | > zα/2 → α
• p-value = Pθ0 [|W | > |w|] ≈ P [|Z| > |w|] = 2Φ(−|w|)
Sample from f (x; θbn ) instead of from Fbn , where θbn could be the mle or method
of moments estimator. Likelihood ratio test (LRT)
15
 supθ∈Θ Ln (θ) Ln (θbn ) Vector parameter
• T (X) = =
supθ∈Θ0 Ln (θ) Ln (θbn,0 ) ( s
X
)
k fX (x | θ) = h(x) exp ηi (θ)Ti (x) − A(θ)
iid
D
X
• λ(X) = 2 log T (X) → χ2r−q where Zi2 ∼ χ2k and Z1 , . . . , Zk ∼ N (0, 1) i=1

 i=1  = h(x) exp {η(θ) · T (x) − A(θ)}

• p-value = Pθ0 [λ(X) > λ(x)] ≈ P χ2r−q > λ(x) = h(x)g(θ) exp {η(θ) · T (x)}
Multinomial LRT Natural form
 
X1 Xk
• mle: pbn = ,..., fX (x | η) = h(x) exp {η · T(x) − A(η)}
n n
k  Xj = h(x)g(η) exp {η · T(x)}
Ln (b
pn ) Y pbj
= h(x)g(η) exp η T T(x)

• T (X) = =
Ln (p0 ) j=1
p0j
k  
X pbj D
15 Bayesian Inference
• λ(X) = 2 Xj log → χ2k−1
j=1
p 0j
Bayes’ Theorem
• The approximate size α LRT rejects H0 when λ(X) ≥ χ2k−1,α
f (x | θ)f (θ) f (x | θ)f (θ)
Pearson Chi-square Test f (θ | x) = =R ∝ Ln (θ)f (θ)
f (xn ) f (x | θ)f (θ) dθ
k
X (Xj − E [Xj ])2 Definitions
• T = where E [Xj ] = np0j under H0
j=1
E [Xj ] • X n = (X1 , . . . , Xn )
D
• T → χ2k−1 • xn = (x1 , . . . , xn )

• p-value = P χ2k−1 > T (x)
 • Prior density f (θ)
D
2 • Likelihood f (xn | θ): joint density of the data
• Faster → Xk−1 than LRT, hence preferable for small n Yn
In particular, X n iid =⇒ f (xn | θ) = f (xi | θ) = Ln (θ)
Independence testing i=1
• Posterior density f (θ | xn )
• I rows, J columns, X multinomial sample of size n = I ∗ J
• Normalizing constant cn = f (xn ) = f (x | θ)f (θ) dθ
R
X
• mles unconstrained: pbij = nij
X • Kernel: part of a density that depends Ron θ
• mles under H0 : pb0ij = pbi· pb·j = Xni· n·j θLn (θ)f (θ)
• Posterior mean θ̄n = θf (θ | xn ) dθ = R Ln (θ)f
R
(θ) dθ
 
PI PJ nX
• LRT: λ = 2 i=1 j=1 Xij log Xi· Xij·j
PI PJ (X −E[X ])2
• PearsonChiSq: T = i=1 j=1 ijE[Xij ]ij 15.1 Credible Intervals
D
• LRT and Pearson → χ2k ν, where ν = (I − 1)(J − 1) Posterior interval
Z b
n
P [θ ∈ (a, b) | x ] = f (θ | xn ) dθ = 1 − α
14 Exponential Family a

Equal-tail credible interval

Scalar parameter Z a Z ∞
n
f (θ | x ) dθ = f (θ | xn ) dθ = α/2
fX (x | θ) = h(x) exp {η(θ)T (x) − A(θ)} −∞ b

= h(x)g(θ) exp {η(θ)T (x)} Highest posterior density (HPD) region Rn

16
 1. P [θ ∈ Rn ] = 1 − α 15.3.1 Conjugate Priors
2. Rn = {θ : f (θ | xn ) > k} for some k Continuous likelihood (subscript c denotes constant)
Likelihood Conjugate prior Posterior hyperparameters
Rn is unimodal =⇒ Rn is an interval

Unif (0, θ) Pareto(xm , k) max x(n) , xm , k + n
Xn
Exp (λ) Gamma (α, β) α + n, β + xi
15.2 Function of parameters i=1
 Pn   
µ0 i=1 xi 1 n
2

2


Let τ = ϕ(θ) and A = {θ : ϕ(θ) ≤ τ }. N µ, σc N µ0 , σ0 + / + 2 ,
σ2 σ2 σ02 σc
Posterior CDF for τ  0 c−1
1 n
+ 2
Z σ02 σc
H(r | xn ) = P [ϕ(θ) ≤ τ | xn ] = f (θ | xn ) dθ

Pn
νσ02 + i=1 (xi − µ)2
A N µc , σ 2 Scaled Inverse Chi- ν + n,
ν+n
square(ν, σ02 )
Posterior density

νλ + nx̄ n
N µ, σ 2 Normal- , ν + n, α + ,
h(τ | xn ) = H 0 (τ | xn ) ν+n 2
scaled Inverse n 2
1X γ(x̄ − λ)
Gamma(λ, ν, α, β) β+ (xi − x̄)2 +
Bayesian delta method 2 i=1 2(n + γ)

−1
Σ−1 −1
Σ−1 −1
  

τ | X n ≈ N ϕ(θ),
b seb ϕ0 (θ)
 b MVN(µ, Σc ) MVN(µ0 , Σ0 ) 0 + nΣc 0 µ0 + nΣ x̄ ,
−1 −1

Σ−1


0 + nΣc
Xn
MVN(µc , Σ) Inverse- n + κ, Ψ + (xi − µc )(xi − µc )T
15.3 Priors Wishart(κ, Ψ) i=1
n
Choice
X xi
Pareto(xmc , k) Gamma (α, β) α + n, β + log
i=1
x mc

• Subjective bayesianism. Pareto(xm , kc ) Pareto(x0 , k0 ) x0 , k0 − kn where k0 > kn

n
• Objective bayesianism.
X
Gamma (αc , β) Gamma (α0 , β0 ) α0 + nαc , β0 + xi
• Robust bayesianism. i=1

Types

• Flat: f (θ) ∝ constant

R∞
• Proper: −∞ f (θ) dθ = 1
R∞
• Improper: −∞ f (θ) dθ = ∞
• Jeffrey’s prior (transformation-invariant):
p p
f (θ) ∝ I(θ) f (θ) ∝ det(I(θ))

• Conjugate: f (θ) and f (θ | xn ) belong to the same parametric family

17
 Discrete likelihood log10 BF10 BF10 evidence
Likelihood Conjugate prior Posterior hyperparameters 0 − 0.5 1 − 1.5 Weak
n n
0.5 − 1 1.5 − 10 Moderate
Bern (p) Beta (α, β) α+
X
xi , β + n −
X
xi 1−2 10 − 100 Strong
i=1 i=1
>2 > 100 Decisive
n n n p
1−p BF 10
p∗ = where p = P [H1 ] and p∗ = P [H1 | xn ]
X X X
Bin (p) Beta (α, β) α+ xi , β + Ni − xi p
1+ 1−p BF10
i=1 i=1 i=1
n
X
NBin (p) Beta (α, β) α + rn, β + xi
i=1
16 Sampling Methods
n
X
Po (λ) Gamma (α, β) α+ xi , β + n 16.1 Inverse Transform Sampling
i=1
Xn Setup
Multinomial(p) Dir (α) α+ x(i)
i=1 • U ∼ Unif (0, 1)
n
X • X∼F
Geo (p) Beta (α, β) α + n, β + xi
i=1
• F −1 (u) = inf{x | F (x) ≥ u}
Algorithm
15.4 Bayesian Testing
1. Generate u ∼ Unif (0, 1)
If H0 : θ ∈ Θ0 : 2. Compute x = F −1 (u)
Z
Prior probability P [H0 ] = f (θ) dθ 16.2 The Bootstrap
Θ0
Z
Let Tn = g(X1 , . . . , Xn ) be a statistic.
Posterior probability P [H0 | xn ] = f (θ | xn ) dθ
Θ0
1. Estimate VF [Tn ] with VFbn [Tn ].
2. Approximate VFbn [Tn ] using simulation:
∗ ∗
Let H0 , . . . , HK−1 be K hypotheses. Suppose θ ∼ f (θ | Hk ), (a) Repeat the following B times to get Tn,1 , . . . , Tn,B , an iid sample from
the sampling distribution implied by Fn b
f (xn | Hk )P [Hk ]
P [Hk | xn ] = PK , i. Sample uniformly X1∗ , . . . , Xn∗ ∼ Fbn .
n
k=1 f (x | Hk )P [Hk ] ii. Compute Tn∗ = g(X1∗ , . . . , Xn∗ ).
Marginal likelihood (b) Then
B B
!2
Z 1 X
∗ 1 X
f (xn | Hi ) = f (xn | θ, Hi )f (θ | Hi ) dθ vboot = V
bb = Tn,b − T∗
Θ
Fn B B r=1 n,r
b=1

Posterior odds (of Hi relative to Hj ) 16.2.1 Bootstrap Confidence Intervals

n n
P [Hi | x ] f (x | Hi ) P [Hi ] Normal-based interval
= ×
P [Hj | xn ] f (xn | Hj ) P [Hj ] Tn ± zα/2 se
b boot
| {z } | {z }
Bayes Factor BFij prior odds Pivotal interval
Bayes factor 1. Location parameter θ = T (F )
18
 2. Pivot Rn = θbn − θ 2. Generate u ∼ Unif (0, 1)
3. Let H(r) = P [Rn ≤ r] be the cdf of Rn Ln (θcand )
∗ ∗
3. Accept θcand if u ≤
4. Let Rn,b = θbn,b − θbn . Approximate H using bootstrap: Ln (θbn )
B
1 X ∗ 16.4 Importance Sampling
H(r)
b = I(Rn,b ≤ r)
B Sample from an importance function g rather than target density h.
b=1
Algorithm to obtain an approximation to E [q(θ) | xn ]:
5. θβ∗ = β sample quantile of (θbn,1
∗ ∗
, . . . , θbn,B ) iid
1. Sample from the prior θ1 , . . . , θn ∼ f (θ)
6. rβ∗ = β sample quantile of (Rn,1
∗ ∗
, . . . , Rn,B ), i.e., rβ∗ = θβ∗ − θbn
Ln (θi )
2. wi = PB ∀i = 1, . . . , B
 
7. Approximate 1 − α confidence interval Cn = â, b̂ where
i=1 Ln (θi )
PB
3. E [q(θ) | xn ] ≈ i=1 q(θi )wi
b −1 1 − α =
 
∗ ∗
â = θbn − H θbn − r1−α/2 = 2θbn − θ1−α/2
2
α
−1 ∗ ∗
b̂ = θbn − Hb
2
= θbn − rα/2 = 2θbn − θα/2 17 Decision Theory
Percentile interval   Definitions
∗ ∗
Cn = θα/2 , θ1−α/2 • Unknown quantity affecting our decision: θ ∈ Θ
• Decision rule: synonymous for an estimator θb
16.3 Rejection Sampling • Action a ∈ A: possible value of the decision rule. In the estimation
context, the action is just an estimate of θ, θ(x).
b
Setup
• Loss function L: consequences of taking action a when true state is θ or
• We can easily sample from g(θ) discrepancy between θ and θ, b L : Θ × A → [−k, ∞).
• We want to sample from h(θ), but it is difficult
Loss functions
k(θ)
• We know h(θ) up to a proportional constant: h(θ) = R • Squared error loss: L(θ, a) = (θ − a)2
k(θ) dθ (
• Envelope condition: we can find M > 0 such that k(θ) ≤ M g(θ) ∀θ K1 (θ − a) a − θ < 0
• Linear loss: L(θ, a) =
K2 (a − θ) a − θ ≥ 0
Algorithm
• Absolute error loss: L(θ, a) = |θ − a| (linear loss with K1 = K2 )
1. Draw θcand ∼ g(θ) • Lp loss: L(θ, a) = |θ − a|p
2. Generate u ∼ Unif (0, 1) (
0 a=θ
k(θcand ) • Zero-one loss: L(θ, a) =
3. Accept θcand if u ≤ 1 a 6= θ
M g(θcand )
4. Repeat until B values of θcand have been accepted
17.1 Risk
Example
Posterior risk
• We can easily sample from the prior g(θ) = f (θ) Z h i
• Target is the posterior h(θ) ∝ k(θ) = f (xn | θ)f (θ) r(θb | x) = L(θ, θ(x))f
b (θ | x) dθ = Eθ|X L(θ, θ(x))
b

• Envelope condition: f (xn | θ) ≤ f (xn | θbn ) = Ln (θbn ) ≡ M

(Frequentist) risk
• Algorithm Z h i
1. Draw θcand ∼ f (θ) R(θ, θ)
b = L(θ, θ(x))f
b (x | θ) dx = EX|θ L(θ, θ(X))
b
19
Bayes risk 18 Linear Regression
ZZ
Definitions
h i
r(f, θ)
b = L(θ, θ(x))f
b (x, θ) dx dθ = Eθ,X L(θ, θ(X))
b
• Response variable Y
• Covariate X (aka predictor variable or feature)
h h ii h i
r(f, θ)
b = Eθ EX|θ L(θ, θ(X)
b = Eθ R(θ, θ)
b

18.1 Simple Linear Regression

h h ii h i
r(f, θ)
b = EX Eθ|X L(θ, θ(X)
b = EX r(θb | X)
Model
17.2 Admissibility Yi = β0 + β1 Xi + i E [i | Xi ] = 0, V [i | Xi ] = σ 2
Fitted line
• θb0 dominates θb if
b0 rb(x) = βb0 + βb1 x
∀θ : R(θ, θ ) ≤ R(θ, θ)
b
Predicted (fitted) values
∃θ : R(θ, θb0 ) < R(θ, θ)
b Ybi = rb(Xi )
• θb is inadmissible if there is at least one other estimator θb0 that dominates Residuals  
it. Otherwise it is called admissible. ˆi = Yi − Ybi = Yi − βb0 + βb1 Xi

Residual sums of squares (rss)

17.3 Bayes Rule
n
X
Bayes rule (or Bayes estimator) rss(βb0 , βb1 ) = ˆ2i
i=1
• r(f, θ)
b = inf e r(f, θ)
θ
e
R Least square estimates
• θ(x)
b = inf r(θb | x) ∀x =⇒ r(f, θ)
b = r(θb | x)f (x) dx
βbT = (βb0 , βb1 )T : min rss
β
b0 ,β
b1
Theorems

• Squared error loss: posterior mean βb0 = Ȳn − βb1 X̄n

Pn Pn
• Absolute error loss: posterior median i=1 (Xi − X̄n )(Yi − Ȳn ) i=1 Xi Yi − nX̄Y
β1 =
b Pn = P n
• Zero-one loss: posterior mode i=1 (Xi − X̄n )
2 2 2
i=1 Xi − nX
 
β0
h i
E βb | X n =
17.4 Minimax Rules β1
σ 2 n−1 ni=1 Xi2 −X n
h i  P 
Maximum risk V βb | X n = 2
R̄(θ)
b = sup R(θ, θ)
b R̄(a) = sup R(θ, a) nsX −X n 1
θ θ r Pn
2
σ i=1 Xi
√
b
Minimax rule se(
b βb0 ) =
sX n n
sup R(θ, θ)
b = inf R̄(θ)
e = inf sup R(θ, θ)
e
θ θe θe θ σ
√
b
se(
b βb1 ) =
sX n
θb = Bayes rule ∧ ∃c : R(θ, θ)
b =c Pn Pn 2
where s2X = n−1 i=1 (Xi − X n )2 and σ b2 = n−21
i=1 
ˆi (unbiased estimate).
Least favorable prior Further properties:
P P
θbf = Bayes rule ∧ R(θ, θbf ) ≤ r(f, θbf ) ∀θ • Consistency: βb0 → β0 and βb1 → β1
20
 • Asymptotic normality: 18.3 Multiple Regression
βb0 − β0 D βb1 − β1 D Y = Xβ + 
→ N (0, 1) and → N (0, 1)
se(
b βb0 ) se(
b βb1 )
where
• Approximate 1 − α confidence intervals for β0 and β1 :      
X11 ··· X1k β1 1
 .. ..  β =  ... 
..  .. 
βb0 ± zα/2 se( and βb1 ± zα/2 se( X= . =.
 
b βb0 ) b βb1 ) . . 
Xn1 ··· Xnk βk n
• Wald test for H0 : β1 = 0 vs. H1 : β1 6= 0: reject H0 if |W | > zα/2 where
W = βb1 /se(
b βb1 ). Likelihood
 
1
R2 L(µ, Σ) = (2πσ 2 )−n/2 exp − 2 rss
Pn b 2
Pn 2 2σ
i=1 (Yi − Y ) ˆ rss
2
R = Pn 2
= 1 − Pn i=1 i 2 = 1 −
i=1 (Yi − Y ) i=1 (Yi − Y )
tss
N
X
Likelihood rss = (y − Xβ)T (y − Xβ) = kY − Xβk2 = (Yi − xTi β)2
n n n i=1
Y Y Y
L= f (Xi , Yi ) = fX (Xi ) × fY |X (Yi | Xi ) = L1 × L2
i=1 i=1 i=1 If the (k × k) matrix X T X is invertible,
Yn
L1 = fX (Xi ) βb = (X T X)−1 X T Y
i=1 h i
V βb | X n = σ 2 (X T X)−1
n
( )
Y 1 X 2
−n
L2 = fY |X (Yi | Xi ) ∝ σ exp − 2 Yi − (β0 − β1 Xi )
2σ i βb ≈ N β, σ 2 (X T X)−1


i=1

Under the assumption of Normality, the least squares parameter estimators are
Estimate regression function
also the MLEs, but the least squares variance estimator is not the MLE
n k
1X 2 X
b2 =
σ ˆ rb(x) = βbj xj
n i=1 i j=1

18.2 Prediction Unbiased estimate for σ 2

Observe X = x∗ of the covariate and want to predict their outcome Y∗ . n
1 X 2
b2 =
σ ˆ ˆ = X βb − Y
Yb∗ = βb0 + βb1 x∗ n − k i=1 i
h i h i h i h i
V Yb∗ = V βb0 + x2∗ V βb1 + 2x∗ Cov βb0 , βb1 mle
n−k 2
Prediction interval µ
b = X̄ b2 =
σ σ
 Pn 2
 n
2 2 i=1 (Xi − X∗ )
ξn = σ
b P +1
n i (Xi − X̄)2 j
b
1 − α Confidence interval
Yb∗ ± zα/2 ξbn βbj ± zα/2 se(
b βbj )
21
18.4 Model Selection Akaike Information Criterion (AIC)
Consider predicting a new observation Y ∗ for covariates X ∗ and let S ⊂ J
denote a subset of the covariates in the model, where |S| = k and |J| = n. bS2 ) − k
AIC(S) = `n (βbS , σ
Issues
Bayesian Information Criterion (BIC)
• Underfitting: too few covariates yields high bias
• Overfitting: too many covariates yields high variance k
bS2 ) − log n
BIC(S) = `n (βbS , σ
Procedure 2

1. Assign a score to each model Validation and training

2. Search through all models to find the one with the highest score
m
X n n
Hypothesis testing R
bV (S) = (Ybi∗ (S) − Yi∗ )2 m = |{validation data}|, often or
i=1
4 2
H0 : βj = 0 vs. H1 : βj 6= 0 ∀j ∈ J
Leave-one-out cross-validation
Mean squared prediction error (mspe)
n n
!2
h i X X Yi − Ybi (S)
mspe = E (Yb (S) − Y ∗ )2 R
bCV (S) = (Yi − Yb(i) )2 =
i=1 i=1
1 − Uii (S)
Prediction risk
n n h i
U (S) = XS (XST XS )−1 XS (“hat matrix”)
X X
R(S) = mspei = E (Ybi (S) − Yi∗ )2
i=1 i=1

Training error
n
R
btr (S) =
X
(Ybi (S) − Yi )2 19 Non-parametric Function Estimation
i=1

R 2 19.1 Density Estimation

Pn b 2
R i=1 (Yi (S) − Y )
rss(S) btr (S) R
R2 (S) = 1 − =1− =1− P n 2
Estimate f (x), where f (x) = P [X ∈ A] = A
f (x) dx.
i=1 (Yi − Y )
tss tss Integrated square error (ise)
The training error is a downward-biased estimate of the prediction risk. Z  Z
2
h i L(f, fbn ) = f (x) − fbn (x) dx = J(h) + f 2 (x) dx
E R btr (S) < R(S)

h i n
X h i Frequentist risk
bias(Rtr (S)) = E Rtr (S) − R(S) = −2
b b Cov Ybi , Yi
i=1 h i Z Z
R(f, fbn ) = E L(f, fbn ) = b2 (x) dx + v(x) dx
Adjusted R2
n − 1 rss
R2 (S) = 1 −
n − k tss
h i
Mallow’s Cp statistic b(x) = E fbn (x) − f (x)
h i
R(S)
b =R σ 2 = lack of fit + complexity penalty
btr (S) + 2kb v(x) = V fbn (x)
22
19.1.1 Histograms KDE
n
Definitions
 
1X1 x − Xi
fbn (x) = K
n i=1 h h
• Number of bins m 1 4
Z
00 2 1
Z
• Binwidth h = m 1 R(f, fn ) ≈ (hσK )
b (f (x)) dx + K 2 (x) dx
4 nh
• Bin Bj has νj observations c
−2/5 −1/5 −1/5
c2 c3
Z Z
h∗ = 1 c = σ 2
, c = K 2
(x) dx, c = (f 00 (x))2 dx
R
• Define pbj = νj /n and pj = Bj f (u) du 1 K 2 3
n1/5
Z 4/5 Z 1/5
∗ c4 5 2 2/5 2 00 2
Histogram estimator R (f, fn ) = 4/5
b c4 = (σK ) K (x) dx (f ) dx
n 4
| {z }
m C(K)
X pbj
fbn (x) = I(x ∈ Bj )
h
j=1 Epanechnikov Kernel
h i pj
E fbn (x) = ( √
h √ 3
|x| < 5
h i p (1 − p ) K(x) = 4 5(1−x2 /5)
j j
V fbn (x) = 0 otherwise
nh2
h2
Z
2 1
R(fbn , f ) ≈ (f 0 (u)) du + Cross-validation estimate of E [J(h)]
12 nh
!1/3
∗ 1 6 n n n  
1 X X ∗ Xi − Xj
Z
h = 1/3 R 2 du 2Xb 2
n (f 0 (u)) JbCV (h) = fbn2 (x) dx − f(−i) (Xi ) ≈ 2
K + K(0)
n i=1 hn i=1 j=1 h nh
 2/3 Z 1/3
∗ b C 3 0 2
R (fn , f ) ≈ 2/3 C= (f (u)) du
n 4 Z
∗ (2) (2)
K (x) = K (x) − 2K(x) K (x) = K(x − y)K(y) dy
Cross-validation estimate of E [J(h)]

Z n m
JbCV (h) = fbn2 (x) dx −
2Xb
f(−i) (Xi ) =
2
−
n+1 X 2
pb
19.2 Non-parametric Regression
n i=1 (n − 1)h (n − 1)h j=1 j
Estimate f (x) where f (x) = E [Y | X = x]. Consider pairs of points
(x1 , Y1 ), . . . , (xn , Yn ) related by

19.1.2 Kernel Density Estimator (KDE) Yi = r(xi ) + i

E [i ] = 0
Kernel K
V [i ] = σ 2
• K(x) ≥ 0
•
R
K(x) dx = 1 k-nearest Neighbor Estimator
R
• xK(x) dx = 0
R 2 2 1 X
• x K(x) dx ≡ σK >0 rb(x) = Yi where Nk (x) = {k values of x1 , . . . , xn closest to x}
k 23
i:xi ∈Nk (x)
Nadaraya-Watson Kernel Estimator 20 Stochastic Processes
n
X
rb(x) = wi (x)Yi Stochastic Process
i=1 (
x−xi {0, ±1, . . . } = Z discrete


K {Xt : t ∈ T } T =
wi (x) = h ∈ [0, 1]
[0, ∞) continuous

Pn x−xj
j=1 K h
4 Z  2
h4 f 0 (x)
Z
2 2 00 0 • Notations Xt , X(t)
rn , r) ≈
R(b x K (x) dx r (x) + 2r (x) dx
4 f (x) • State space X
Z 2
σ K 2 (x) dx
R
• Index set T
+ dx
nhf (x)
c1
h∗ ≈ 1/5 20.1 Markov Chains
n
∗ c2
rn , r) ≈ 4/5
R (b Markov chain
n
P [Xn = x | X0 , . . . , Xn−1 ] = P [Xn = x | Xn−1 ] ∀n ∈ T, x ∈ X
Cross-validation estimate of E [J(h)]
n n Transition probabilities
X X (Yi − rb(xi ))2
JbCV (h) = (Yi − rb(−i) (xi ))2 = !2
i=1 i=1 K(0)
pij ≡ P [Xn+1 = j | Xn = i]
1−  x−x 
Pn
j=1 K h
j
pij (n) ≡ P [Xm+n = j | Xm = i] n-step

19.3 Smoothing Using Orthogonal Functions Transition matrix P (n-step: Pn )

Approximation • (i, j) element is pij

∞ J
X X • pij > 0
r(x) = βj φj (x) ≈ βj φj (x) P
j=1 i=1
• i pij = 1

Multivariate regression Chapman-Kolmogorov

Y = Φβ + η
  X
φ0 (x1 ) ··· φJ (x1 ) pij (m + n) = pij (m)pkj (n)
 .. .. .. 
where ηi = i and Φ =  . . . 
k

φ0 (xn ) · · · φJ (xn )
Pm+n = Pm Pn
Least squares estimator
βb = (ΦT Φ)−1 ΦT Y Pn = P × · · · × P = Pn
1
≈ ΦT Y (for equally spaced observations only) Marginal probability
n
Cross-validation estimate of E [J(h)] µn = (µn (1), . . . , µn (N )) where µi (i) = P [Xn = i]
2
µ0 , initial distribution

Xn J
X
R
bCV (J) = Yi − φj (xi )βbj,(−i)  µn = µ0 Pn
24
i=1 j=1
20.2 Poisson Processes Autocorrelation function (ACF)
Poisson process
Cov [xs , xt ] γ(s, t)
ρ(s, t) = p =p
• {Xt : t ∈ [0, ∞)} = number of events up to and including time t V [xs ] V [xt ] γ(s, s)γ(t, t)
• X0 = 0
• Independent increments: Cross-covariance function (CCV)
∀t0 < · · · < tn : Xt1 − Xt0 ⊥
⊥ · · · ⊥⊥ Xtn − Xtn−1
γxy (s, t) = E [(xs − µxs )(yt − µyt )]
• Intensity function λ(t)
– P [Xt+h − Xt = 1] = λ(t)h + o(h) Cross-correlation function (CCF)
– P [Xt+h − Xt = 2] = o(h)
Rt γxy (s, t)
• Xs+t − Xs ∼ Po (m(s + t) − m(s)) where m(t) = 0
λ(s) ds ρxy (s, t) = p
γx (s, s)γy (t, t)
Homogeneous Poisson process
Backshift operator
λ(t) ≡ λ =⇒ Xt ∼ Po (λt) λ>0
B k (xt ) = xt−k
Waiting times
Wt := time at which Xt occurs
  Difference operator
1
Wt ∼ Gamma t, ∇d = (1 − B)d
λ
Interarrival times
White noise
St = Wt+1 − Wt
 
1 2
• wt ∼ wn(0, σw )
St ∼ Exp
λ iid 2


• Gaussian: wt ∼ N 0, σw
St
• E [wt ] = 0 t ∈ T
• V [wt ] = σ 2 t ∈ T
Wt−1 Wt t • γw (s, t) = 0 s 6= t ∧ s, t ∈ T

Random walk
21 Time Series
• Drift δ
Mean function ∞
Pt
• xt = δt + j=1 wj
Z
µxt = E [xt ] = xft (x) dx
−∞ • E [xt ] = δt
Autocovariance function
Symmetric moving average
γx (s, t) = E [(xs − µs )(xt − µt )] = E [xs xt ] − µs µt
k
X k
X
γx (t, t) = E (xt − µt )2 = V [xt ]
 
mt = aj xt−j where aj = a−j ≥ 0 and aj = 1
25
j=−k j=−k
21.1 Stationary Time Series 21.2 Estimation of Correlation
Strictly stationary Sample mean
n
1X
x̄ = xt
P [xt1 ≤ c1 , . . . , xtk ≤ ck ] = P [xt1 +h ≤ c1 , . . . , xtk +h ≤ ck ] n t=1

Sample variance
n 
∀k ∈ N, tk , ck , h ∈ Z

1 X |h|
V [x̄] = 1− γx (h)
n n
h=−n
Weakly stationary
  Sample autocovariance function
• E x2t < ∞ ∀t ∈ Z
n−h
1 X
 2
• E xt = m ∀t ∈ Z γ
b(h) = (xt+h − x̄)(xt − x̄)
• γx (s, t) = γx (s + r, t + r) ∀r, s, t ∈ Z n t=1

Autocovariance function Sample autocorrelation function

γ
b(h)
• γ(h) = E [(xt+h − µ)(xt − µ)] ∀h ∈ Z ρb(h) =
  γ
b(0)
• γ(0) = E (xt − µ)2
• γ(0) ≥ 0 Sample cross-variance function
• γ(0) ≥ |γ(h)|
n−h
• γ(h) = γ(−h) 1 X
γ
bxy (h) = (xt+h − x̄)(yt − y)
n t=1
Autocorrelation function (ACF)
Sample cross-correlation function
Cov [xt+h , xt ] γ(t + h, t) γ(h)
ρx (h) = p =p = γ
bxy (h)
V [xt+h ] V [xt ] γ(t + h, t + h)γ(t, t) γ(0) ρbxy (h) = p
γbx (0)b
γy (0)
Jointly stationary time series Properties

γxy (h) = E [(xt+h − µx )(yt − µy )] 1

• σρbx (h) = √ if xt is white noise
n
1
γxy (h) • σρbxy (h) = √ if xt or yt is white noise
ρxy (h) = p n
γx (0)γy (h)
21.3 Non-Stationary Time Series
Linear process
Classical decomposition model
∞
X ∞
X
xt = µ + ψj wt−j where |ψj | < ∞ xt = µt + st + wt
j=−∞ j=−∞
• µt = trend
∞
X • st = seasonal component
2
γ(h) = σw ψj+h ψj • wt = random noise term
26
j=−∞
21.3.1 Detrending Moving average polynomial
Least squares θ(z) = 1 + θ1 z + · · · + θq zq z ∈ C ∧ θq 6= 0

1. Choose trend model, e.g., µt = β0 + β1 t + β2 t2 Moving average operator

2. Minimize rss to obtain trend estimate µ bt = βb0 + βb1 t + βb2 t2 θ(B) = 1 + θ1 B + · · · + θp B p
3. Residuals , noise wt
MA (q) (moving average model order q)
Moving average xt = wt + θ1 wt−1 + · · · + θq wt−q ⇐⇒ xt = θ(B)wt
1
• The low-pass filter vt is a symmetric moving average mt with aj = 2k+1 :
q
X
E [xt ] = θj E [wt−j ] = 0
k j=0
1 X
vt = xt−1 (
2
Pq−h
2k + 1 σw j=0 θj θj+h 0≤h≤q
i=−k γ(h) = Cov [xt+h , xt ] =
0 h>q
1
Pk
• If 2k+1 i=−k wt−j ≈ 0, a linear trend function µt = β0 + β1 t passes MA (1)
without distortion xt = wt + θwt−1

2 2
Differencing (1 + θ )σw h = 0

γ(h) = θσw 2
h=1
• µt = β0 + β1 t =⇒ ∇xt = β1 
0 h>1

(
θ
21.4 ARIMA models 2 h=1
ρ(h) = (1+θ )
Autoregressive polynomial 0 h>1
ARMA (p, q)
φ(z) = 1 − φ1 z − · · · − φp zp z ∈ C ∧ φp 6= 0
xt = φ1 xt−1 + · · · + φp xt−p + wt + θ1 wt−1 + · · · + θq wt−q
Autoregressive operator φ(B)xt = θ(B)wt
p Partial autocorrelation function (PACF)
φ(B) = 1 − φ1 B − · · · − φp B
• xih−1 , regression of xi on {xh−1 , xh−2 , . . . , x1 }
Autoregressive model order p, AR (p)
• φhh = corr(xh − xh−1
h , x0 − xh−1
0 ) h≥2
xt = φ1 xt−1 + · · · + φp xt−p + wt ⇐⇒ φ(B)xt = wt • E.g., φ11 = corr(x1 , x0 ) = ρ(1)
ARIMA (p, d, q)
AR (1)
∇d xt = (1 − B)d xt is ARMA (p, q)
k−1
X k→∞,|φ|<1
∞
X φ(B)(1 − B)d xt = θ(B)wt
• xt = φk (xt−k ) + φj (wt−j ) = φj (wt−j )
Exponentially Weighted Moving Average (EWMA)
j=0 j=0
xt = xt−1 + wt − λwt−1
| {z }
linear process
P∞ j ∞
• E [xt ] = j=0 φ (E [wt−j ]) = 0 X
2 h
σw φ
xt = (1 − λ)λj−1 xt−j + wt when |λ| < 1
• γ(h) = Cov [xt+h , xt ] = 1−φ2 j=1
γ(h)
• ρ(h) = γ(0) = φh x̃n+1 = (1 − λ)xn + λx̃n
• ρ(h) = φρ(h − 1) h = 1, 2, . . . Seasonal ARIMA
27
 • Denoted by ARIMA (p, d, q) × (P, D, Q)s Periodic mixture
• ΦP (B s )φ(B)∇D d s
s ∇ xt = δ + ΘQ (B )θ(B)wt q
X
xt = (Uk1 cos(2πωk t) + Uk2 sin(2πωk t))
21.4.1 Causality and Invertibility k=1
P∞
ARMA (p, q) is causal (future-independent) ⇐⇒ ∃{ψj } : j=0 ψj < ∞ such that • Uk1 , Uk2 , for k = 1, . . . , q, are independent zero-mean rv’s with variances σk2
Pq
∞
• γ(h) = k=1 σk2 cos(2πωk h)
  Pq
• γ(0) = E x2t = k=1 σk2
X
xt = wt−j = ψ(B)wt
j=0
Spectral representation of a periodic process
P∞
ARMA (p, q) is invertible ⇐⇒ ∃{πj } : πj < ∞ such that
j=0 γ(h) = σ 2 cos(2πω0 h)
∞ σ 2 −2πiω0 h σ 2 2πiω0 h
= e + e
X
π(B)xt = Xt−j = wt 2 2
j=0 Z 1/2
= e2πiωh dF (ω)
Properties −1/2

• ARMA (p, q) causal ⇐⇒ roots of φ(z) lie outside the unit circle Spectral distribution function
∞

X θ(z)j 0
 ω < −ω0
ψ(z) = ψj z = |z| ≤ 1
φ(z) F (ω) = σ 2 /2 −ω ≤ ω < ω0
j=0 
 2
σ ω ≥ ω0
• ARMA (p, q) invertible ⇐⇒ roots of θ(z) lie outside the unit circle
• F (−∞) = F (−1/2) = 0
∞
X φ(z) • F (∞) = F (1/2) = γ(0)
π(z) = πj z j = |z| ≤ 1
θ(z)
j=0 Spectral density
Behavior of the ACF and PACF for causal and invertible ARMA models ∞
X 1 1
f (ω) = γ(h)e−2πiωh − ≤ω≤
AR (p) MA (q) ARMA (p, q) 2 2
h=−∞
ACF tails off cuts off after lag q tails off
P∞ R 1/2
PACF cuts off after lag p tails off q tails off • Needs h=−∞ |γ(h)| < ∞ =⇒ γ(h) = −1/2
e2πiωh f (ω) dω h = 0, ±1, . . .
• f (ω) ≥ 0
21.5 Spectral Analysis • f (ω) = f (−ω)
Periodic process • f (ω) = f (1 − ω)
R 1/2
• γ(0) = V [xt ] = −1/2 f (ω) dω
xt = A cos(2πωt + φ) 2
• White noise: fw (ω) = σw
= U1 cos(2πωt) + U2 sin(2πωt)
• ARMA (p, q) , φ(B)xt = θ(B)wt :
• Frequency index ω (cycles per unit time), period 1/ω |θ(e−2πiω )|2
2
• Amplitude A fx (ω) = σw
|φ(e−2πiω )|2
• Phase φ
Pp Pq
• U1 = A cos φ and U2 = A sin φ often normally distributed rv’s where φ(z) = 1 − k=1 φk z k and θ(z) = 1 + k=1 θk z k
28
Discrete Fourier Transform (DFT) • I0 (a, b) = 0 I1 (a, b) = 1
n • Ix (a, b) = 1 − I1−x (b, a)
X
d(ωj ) = n−1/2 xt e−2πiωj t
i=1 22.3 Series
Fourier/Fundamental frequencies Finite Binomial
n n  
ωj = j/n X n(n + 1) X n
• k= • = 2n
2 k
Inverse DFT k=1 k=0
n−1 n n    
X r+k r+n+1
xt = n−1/2 d(ωj )e2πiωj t
X X
• (2k − 1) = n2 • =
j=0
k n
k=1 k=0
n n    
Periodogram X n(n + 1)(2n + 1) X k n+1
2 • k2 = • =
I(j/n) = |d(j/n)| 6 m m+1
k=1 k=0
Scaled Periodogram n  2 • Vandermonde’s Identity:
X n(n + 1)
• k3 = r  
m n
 
m+n

2
X
4 k=1 =
P (j/n) = I(j/n) n k r−k r
n X cn+1 − 1 k=0
n
!2 n
!2 • ck = c 6= 1 • Binomial Theorem:
2X 2X c−1 n  
n n−k k
= xt cos(2πtj/n + xt sin(2πtj/n k=0
X
n t=1 n t=1 a b = (a + b)n
k
k=0

22 Math Infinite
∞ ∞
22.1 Gamma Function
X 1 X p
• pk = , pk = |p| < 1
Z ∞ 1−p 1−p
k=0 k=1
• Ordinary: Γ(s) = ts−1 e−t dt ∞ ∞
!  
0
X d X d 1 1
Z ∞ • kpk−1 = pk
= = |p| < 1
• Upper incomplete: Γ(s, x) = ts−1 e−t dt dp dp 1 − p (1 − p)2
k=0 k=0
x ∞
X r + k − 1

Z x
• Lower incomplete: γ(s, x) = ts−1 e−t dt • xk = (1 − x)−r r ∈ N+
k
0 k=0
∞  
• Γ(α + 1) = αΓ(α) α>1 X α k
• p = (1 + p)α |p| < 1 , α ∈ C
• Γ(n) = (n − 1)! n∈N k
√ k=0
• Γ(1/2) = π

22.2 Beta Function

Z1
Γ(x)Γ(y)
• Ordinary: B(x, y) = B(y, x) = tx−1 (1 − t)y−1 dt =
0 Γ(x + y)
Z x
• Incomplete: B(x; a, b) = ta−1 (1 − t)b−1 dt
0
• Regularized incomplete:
a+b−1
B(x; a, b) a,b∈N X (a + b − 1)!
Ix (a, b) = = xj (1 − x)a+b−1−j
B(a, b) j=a
j!(a + b − 1 − j)! 29
22.4 Combinatorics
Sampling

k out of n w/o replacement w/ replacement

k−1
Y n!
ordered nk = (n − i) = nk
i=0
(n − k)!
nk
     
n n! n−1+r n−1+r
unordered = = =
k k! k!(n − k)! r n−1

Stirling numbers, 2nd kind

        (
n n−1 n−1 n 1 n=0
=k + 1≤k≤n =
k k k−1 0 0 else

Partitions
n
X
Pn+k,k = Pn,i k > n : Pn,k = 0 n ≥ 1 : Pn,0 = 0, P0,0 = 1
i=1

Balls and Urns f :B→U D = distinguishable, ¬D = indistinguishable.

|B| = n, |U | = m f arbitrary f injective f surjective f bijective

( (
mn m ≥ n
 
n n! m = n
B : D, U : D mn m!
0 else m 0 else
      (
m+n−1 m n−1 1 m=n
B : ¬D, U : D
n n m−1 0 else
m  
(   (
X n 1 m≥n n 1 m=n
B : D, U : ¬D
k 0 else m 0 else
k=1
m
( (
X 1 m≥n 1 m=n
B : ¬D, U : ¬D Pn,k Pn,m
k=1
0 else 0 else

References
[1] L. M. Leemis and J. T. McQueston. Univariate Distribution Relationships. The American
Statistician, 62(1):45–53, 2008.
[2] A. Steger. Diskrete Strukturen – Band 1: Kombinatorik, Graphentheorie, Algebra.
Springer, 2001.
[3] A. Steger. Diskrete Strukturen – Band 2: Wahrscheinlichkeitstheorie und Statistik.
Springer, 2002.
30
 Univariate distribution relationships, courtesy Leemis and McQueston [1].
31