Time Series Analysis

Sngmng·2023년 2월 18일
0
post-custom-banner

Introduction to Time Series and Forecasting
Peter J. Brockwell, Richard A. Davis 를 바탕으로 작성되었습니다.

Definition of time series model

A General Approach to time series modelling

Autocovariance & Autocorrelation

Esitmation and Elimination of Trend and seasonality



Estimation and Elimination of Both Trend and seasonality


Autoregressive model (AR)

https://en.wikipedia.org/wiki/Autoregressive_model

자기자신을 통한 예측과정이라고 할 수 있고 AR(p)는 다음과 같이 정의된다.

Xt=i=1pϕiXti+ZtX_t = \sum_{i=1}^{p}\phi_iX_{t-i} +Z_t

Zt:Z_t : white noise
혹은 이와 동등하게 다음과 같이 정의된다.

Xt=i=1pϕiBiXt+ZtX_t = \sum_{i=1}^{p}\phi_iB^iX_{t} +Z_t

Bi:B^i: back shift operator (lag operator)

stationary condition은 characteristic equation을 통해서 구해진다.
https://m.blog.naver.com/PostView.naver?isHttpsRedirect=true&blogId=jahyone20&logNo=220933491078

Moving Average model (MA)

https://en.wikipedia.org/wiki/Moving-average_model

MA(q)는 다음과 같이 정의된다.

Xt=μ+i=1qθiZti+ZtX_t = \mu + \sum_{i=1}^{q}\theta_iZ_{t-i}+Z_t

ZZ : white noise error term
μ\mu : mean

back shift operator를 이용한 정의는 다음과 같다.

Xt=μ+(1+θ1B++θqBq)ZtX_t = \mu + (1 +\theta_1B + \cdots + \theta_qB^q)Z_t

Autoregressive moving average model(ARMA)

Def of ARMA(p,q)

Xt=i=1qθiZti+Zt+i=1pϕiXtiX_t = \sum_{i=1}^{q}\theta_iZ_{t-i}+Z_t + \sum_{i=1}^{p}\phi_iX_{t-i}

Case of ARMA(1,1)

XtϕXt1=Zt+θZt1X_t - \phi X_{t-1} = Z_t + \theta Z_{t-1}
where {Zt}\{Z_t\} ~ WN(0,σ2)WN(0,\sigma ^2) and ϕ+θ0\phi +\theta \neq 0

or

ϕ(B)Xt=θ(B)Zt\phi(B)X_t = \theta(B)Z_t

Existence of stationary solution



invertible / noninvertible


Yule-Walker Estimation

Yule-Walker Equation을 통해 time series model의 parameter를 추정하는 방법이다.
AR model의 경우는 다음과 같다.

Bartlett’s Formula

...

Multivariate ARMA

...

State-Space Representations

State-Space model은 아래 두개의 방정식으로 구성되는
time series {Yt\{\bf{Y}_t, t=1,2,}t = 1,2,\dots\} 를 말한다.

Yt=GtXt+Wt{\bf Y}_t =G_t{\bf X}_t+{\bf W}_t, t=1,2,t=1, 2, \dots : observation equation

Xt+1=FtXt+Vt{\bf X}_{t+1} =F_t{\bf X}_t+{\bf V}_t, t=1,2,t=1, 2, \dots : state equation

where
{Wt}WN(0,{Rt})\{{\bf W}_t\} \sim WN({\bf 0},\{R_t \})
, {Vt}WN(0,{Qt})\{{\bf V}_t\} \sim WN({\bf 0},\{Q_t \})
, Gt{G_t} is sequence of ww x vv matrices
, Ft{F_t} is sequence of vv x vv matrices
, Xt{\bf X}_t is vv-dimensional state variable
, Yt{\bf Y}_t is ww-dimensional observation
and VtV_t is uncorrelated with WtW_t ( i.e E(WtVtT)=0E(W_tV_t^T) = 0 for all )

ARMA(1,1) process

profile
개인 공부 기록용
post-custom-banner

0개의 댓글