-
Notifications
You must be signed in to change notification settings - Fork 18
/
VI.jl
102 lines (83 loc) · 2.62 KB
/
VI.jl
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
## Bayesian analysis of the steady-state AR model with one lag
## Loading modules and settings
using Plots, Statistics, Distributions
using LaTeXStrings
using DelimitedFiles
using ColorSchemes
import ColorSchemes: Paired_12; colors = Paired_12
basicColors = Paired_12[[1,2,3,4,7,8]]
gr(legend = nothing)
## Setting up functions
"""
Simulate a AR(1) with steady-state
"""
function SimSteadyStateAR(nTime, μ, ϕ, σ, nLags = 1)
if nLags != 1 error("Only NLags = 1 is implemented. For teaching") end
y = zeros(nTime+1)
for t in 1:nTime
y[t+1] = μ + ϕ*(y[t]-μ) + σ*randn()
end
return(y[2:end])
end
ScaledInverseChiSq(ν,τ²) = InverseGamma(ν/2,ν*τ²/2)
"""
Gibbs sampling for the steady state AR(1) model
Samples `nIter` draws from the posterior of the model:
```math
yₜ = μ + ϕ(yₜ₋₁ - μ) + ε, where ε ∼ N(0,σ²)
```
with prior
```math
μ ∼ N(0,ψ)
ϕ | σ² ∼ N(0,σ²/κ₀)
σ² ∼ Inv-χ²(ν₀,σ₀^2)
```
"""
function GibbsSteadyStateAR(y, nIter, ψ, κ₀, ν₀, σ₀)
# Set up storage and prelims
n = length(y)
postDraws = zeros(nIter,3)
# Initial values
x = [0;y[1:end-1]] # Set up the lag
ϕ = x'x \ x'y
μ = mean(y)
σ² = var(y - ϕ*x)
for i in 1:nIter
# Tranforming the data a bit
yTilde = y .- μ
xTilde = [0;yTilde[1:end-1]] # Set up the lag
# Draw ϕ
ϕTilde = (xTilde'xTilde + κ₀) \ (x'yTilde)
ϕ = rand(Normal(ϕTilde,√(σ²/(xTilde'xTilde + κ₀))))
postDraws[i,1] = ϕ
# Draw σ²
νₙ = ν₀ + n
τ² = ( (yTilde - ϕ*xTilde)'*(yTilde - ϕ*xTilde) + κ₀*ϕ^2 + ν₀*σ₀^2)/νₙ
σ² = rand(ScaledInverseChiSq(νₙ,τ²))
postDraws[i,2] = σ²
# Draw μ
yCheck = (y - ϕ*x)/(1 - ϕ)
w = ((n/σ²)*(1-ϕ)^2)/((n/σ²)*(1-ϕ)^2 + 1/ψ^2)
τₙ = √( 1/( (n/σ²)*(1-ϕ)^2 + 1/ψ^2 ) )
μ = rand(Normal(w*mean(yCheck),τₙ))
postDraws[i,3] = μ
end
return(postDraws)
end
## Simulating data and sampling from posterior
# Simulate data
μ = 1; ϕ = 0.5; σ = 1; n = 200;
y = SimSteadyStateAR(n, μ, ϕ, σ)
plot(y, lw = 1.5, xlab = "time", ylab = "y", color = basicColors[6]);
writedlm("timeseries.csv", y, ' ')
z = readdlm("/home/mv/Dropbox/Teaching/AdvBayesLearnCourse/Labs/timeseries.csv")
# Set prior
ψ = 1; κ₀ = 8; ν₀ = 4; σ₀ = 1;
# Sample from posterior
nIter = 10000
postDraws = GibbsSteadyStateAR(y, nIter, ψ, κ₀, ν₀, σ₀)
mean(postDraws, dims = 1)
p1 = histogram(postDraws[:,1]);
p2 = histogram(postDraws[:,2]);
p3 = histogram(postDraws[:,3]);
plot(p1,p2,p3)