Analysis using R for seasonal adjustment and trend estimate. - - PowerPoint PPT Presentation

Web Decomp and

E-decomp -Time Series

Analysis using R

Seisho Sato

（ The Institute of Statistical Mathematics ） http://www.ism.ac.jp/~sato/

Decomp

■ originally developed by Kitagawa(1984). ■ for seasonal adjustment and trend estimate. ■ written by Fortran. ■ using Square Root Filter.

Decomp family

■ Timsac84 (Fortran) ■ Tsview.uni (S) ■ Web Decomp (CGI & S or R) ■ S-Decomp (S) ■ MITI-Decomp (S & Excel) ■ E-Decomp (Excel VBA & R)

R-(D)COM interface (by Thomas Baier)

Concepts of “Web Decomp”

◆ A WWW Site ◆ Statistical Software of Time Series model

including Seasonal Adjustment model (Decomp)

◆ All Calculation can be done by server

machine

Client Computer (PC, MAC, Workstation) Server Machine （ISM） The Internet Send Data WWW Browser (Netscape Navigator

• r Internet Explore)

WWW Server

Fig 1. System of Web Decomp

CGI Program （Interface for Web Decomp） S Language （Graphics） Fortran Program （Numerical Calculation） History of Analysis HTML Documents Simple Calculation by using JavaScript R eci eve R esul t R or S

Menu of “Web Decomp”

■

Decomp ―――― State space modeling of seasonal adjustment (Kitagwa and Gersh 1984)

■

plot ―――― Time series plot

■

autocor ――― Plot of autocorrelation (TIMSAC72)

■

spectrum ――― Non-parametric spectrum (TIMSAC72)

■

ARfit ―――― Fitting AR model (TIMSAC72)

■

ARMAfit ―――― Fitting ARMA model （ Kitagawa(1993) ）

■

log ―――― Log-transformation

Menu of “Web Decomp”

◆ diff ――――

First difference

◆ diff4 ――――

Seasonal difference for quarterly data

◆ diff12 ――――

Seasonal difference for monthly data

◆ Volatility ――――

Fitting Stochastic Volatility model

(More methods will be added)

Windows of Web Decomp

access numbers of Web Decomp

access num ber ( m onthl y) 200 400 600 800 1000 1200 1400 1600 1800 Dec-04 Jan-05 Feb-05 M ar-05 Apr-05 M ay-05 Jun-05 Jul

• 05

Aug-05 Sep-05 O ct-05 Nov-05 Dec-05 Jan-06 Feb-06 M ar-06 access num ber

Total = 18871

About Decomp

■ State Space Model

Δd T(t) = e1(t) S(t) = -S(t-1) - … -S(t-p) + e2(t) A(t) = a1 A(t-1) + … + aq A(t-q) + e3(t) y(t) = T(t) + S(t) + A(t) + TD(t) + e4(t) T: Trend S: Seasonal A: AR (Cyclical component) TD: Trading day y: Observation e1-e4: i.i.d noise

Parameter of Decomp

Log transform: Seasonal frequency: Trend Order: AR Order: Trading Day Effects:

Output

■ Graph output ■ Data output ■ Value of parameter and other

statistics

*. AC. JP （ Japanese Educat i

• n)

1414 *. O R. JP （ Pr

• vi

der et c. ） 786 *. NE. JP （ Pr

• vi

der , O CN et c） 759 *. CO . JP （ Japanese com pany） 726 *. CO M （ US com pany） 284 *. GO . JP （ Japanese gover m ent ） 171 *. AD. JP （ Pr

• vi

der , M esh et c） 152 *. NET （ US） 66 *. EDU （ US Educat i

• n）

53 *. TW （ Tai wan） 23 *. HR （ Cr

• at

i a） 17 *. DE （ Ger m any） 17 *. UK, *. GB （ UK） 14 *. TO KYO . JP （ Tokyo M et r

• ）

14 （ unknown） 1439

Summary of access list (Jun. 1997 - Jan. 1998)

E-Decomp

EXCEL VBA R R-(D)COM DLL E-Decomp