Evaluation of X11 & Model-based Methods of Seasonal Adjustment - - PowerPoint PPT Presentation

Conference on Seasonality, Seasonal Adjustment and their implications for Short-Term Analysis and Forecasting

10-12 May 2006

Evaluation of X11 & Model-based Methods of Seasonal Adjustment for Economic Time Series

Stuart Scott Stuart Scott

Evaluation of X11 & Model-based Methods of Seasonal Adjustment for Economic Time Series

Stuart Scott US Bureau of Labor Statistics Eurostat Conference on Seasonality 10-12 May 2006

“work in progress” – seek your feedback project aims (1) to lead BLS to consider greater use of models (2) to serve the research community with an up-to-date evaluation of methods

Outline

• project management & experimental design
• diagnostics
• results

Seasonal adjustment at BLS: decentralized

• technical work, production runs carried out

independently by each program

• some coordination achieved by me in small

central research office

• method: augmented X11

Exception: bivariate models for state labor force series

Project management – “a team effort”

• bureau-wide team
• responsible to Innovation Board

team products, quarterly progress reports

• Jan 05 start

Initial Stages

• education
• current practices
• computing

Scope of Phase 1 Program # of series CES Current Employment Statistics 25 CPI Consumer Price Index 35 PPI Producer Price Index 22

current status initial runs complete, including summary tables evaluation of initial runs in progress evaluation report on Phase 1 due in July

“inefficient but fruitful” exposure of practitioners to methods in greater depth joint effort to learn more about model- based methods & diagnostics expansion of BLS seasonal adjustment community (of 16 involved, 5 new to SA, at least 2 or 3 likely to work in area)

Evaluation Design – Phase 1

• a. “Automatic” runs with X13 software

methods: X11 ARIMA (using SEATS spec in X13) automatic choices: mode (mult. vs. add.), model, outliers

Evaluation Design – Phase 1

• b. Analyst selection of models, options, outliers

comparison of model selection: TRAMO, X13’s AUTOMDL & PICKMDL methods: X11 ARIMA structural – STAMP, SSMB (Jain, 2001)

Phase 2 – Accounting for sampling error

methods X11 ARIMA or structural models with sampling error component evaluation impact on seasonal adjustment significance of monthly change

impact on seasonal adjustment US state labor force statistics Tiller (1992) time series estimates as small domain estimation technique Tiller (2006) seasonal adjustment from bivariate seasonal models

variance measures for seasonally adjusted series model-based X11 (Pfeffermann, 1994; Scott, Sverchkov, & Pfeffermann, 2005)

Diagnostics

X11 Quality Control statistics Lothian & Morry (1978) models Ljung-Box, AIC, normality

Granger (1978) The criteria I suggested have been shown to be impossible to achieve in practice, and, thus, should be replaced by achievable

• criteria. However, I am at a loss to know

what these criteria should be.

Cross-methods diagnostics spectra sliding spans, revisions monthplot (& “overall F”) decomposition of change in the observed series

presence/absence of seasonality from spectra differenced observed series differenced seasonally adjusted series irregular model residuals graphs (X13’s “6-star method)

Stability

sliding spans 60th %-ile & max of month-month change in seasonally adjusted series revisions 75th %-ile & max of revisions (“final” based on at least 2 years beyond “initial”)

relative contribution of components to change in the observed series

1

1 1

n t d t d t d

X X n d X

= + −

= − −

∑

• mult. case

'2 2 2 2 2 '2

, , , and 1 ,3,12,24 100% /

d d d d d d

X O T S I d O T S I S O = = = + + ×

• rel. contribution
• f seasonal

comparison of ARIMA model parameters and X11 filter choices Depoutot & Planas (1998) Chu, Tiao, & Bell (2006)

Overall approach

• assess presence of seasonality in the series
• assess each method

acceptable, questionable, unacceptable

• compare methods

Results – Phase 1 “automatic” runs

early personal impressions examples of cross-method diagnostics illustration of features of X13 software

Series and codes for PPI evaluation Commodity group/code Series 02 – Processed Foods & Feeds 0221 PMEAT Meats (IA) 022101 PBEEF Beef & Veal 022103 PLAMB Lamb/Mutton 022104 PPORK Pork products 022105 PMOTH Other meats 05 – Fuels, etc 057103 PGASP Unleaded premium gasoline 057302 POILH Home heating oil 057303 PDIE2 #2 diesel fuel 09 – Pulp & Paper 093 PPUBL Publication & printed matter

TD TD TD Meats Autoregressive Spectrum (Decibels), PPI, SEATS

Unadjusted Seasonally Adjusted Visually Significant Peak Visually Significant Peak Unadjusted Median Seasonally Adjusted Median

• 41
• 40
• 39
• 38
• 37
• 36
• 35
• 34
• 33
• 32
• 31
• 30
• 29
• 28
• 27
• 26
• 25

Period in Months 12 6 4 3 2.4 2

Meats Monthplot, PPI, SEATS

Seasonal Means Seasonal Factors 0.97 0.98 0.99 1.00 1.01 1.02 1.03

Jan Feb Mar Apr May Jun Jul Aug Sep Oct Nov Dec

Meats Monthplot, PPI, X11

Seasonal Means Seasonal Factors 0.97 0.98 0.99 1.00 1.01 1.02 1.03 1.04

Jan Feb Mar Apr May Jun Jul Aug Sep Oct Nov Dec

TD TD TD Publications Autoregressive Spectrum (Decibels), PPI, SEATS

Unadjusted Seasonally Adjusted Visually Significant Peak Visually Significant Peak Unadjusted Median Seasonally Adjusted Median

• 20
• 10

10

Period in Months 12 6 4 3 2.4 2

TD TD TD Publications Autoregressive Spectrum (Decibels), PPI, X11

Unadjusted Seasonally Adjusted Visually Significant Peak Visually Significant Peak Unadjusted Median Seasonally Adjusted Median

• 30
• 20
• 10

10

Period in Months 12 6 4 3 2.4 2

Publications Monthplot, PPI, X11

Seasonal Means Seasonal Factors

• 0.8
• 0.7
• 0.6
• 0.5
• 0.4
• 0.3
• 0.2
• 0.1

0.0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9

Jan Feb Mar Apr May Jun Jul Aug Sep Oct Nov Dec

Scenic Trans, CES, SEATS

Series scentran NBER Recessions in Gray

Unadjusted Seasonally Adjusted Trend

19000 20000 21000 22000 23000 24000 25000 26000 27000 28000 29000 30000 31000 32000 33000 34000 35000 36000

1-95 1-96 1-97 1-98 1-99 1-00 1-01 1-02 1-03 1-04 1-05

19000 20000 21000 22000 23000 24000 25000 26000 27000 28000 29000 30000 31000 32000 33000 34000 35000 36000

Scenic Trans Monthplot, CES, SEATS

Seasonal Means Seasonal Factors 0.7 0.8 0.9 1.0 1.1 1.2 1.3 1.4

Jan Feb Mar Apr May Jun Jul Aug Sep Oct Nov Dec

SCENTRAN

Decomposition statistics (%) Trend Seasonal Irregular X11 2.4 91.2 6.4 SEATS 0.8 97.9 1.3

TD TD TD #2 Diesel Autoregressive Spectrum (Decibels), PPI, X11

Unadjusted Seasonally Adjusted Visually Significant Peak Visually Significant Peak Unadjusted Median Seasonally Adjusted Median

• 31
• 30
• 29
• 28
• 27
• 26
• 25
• 24
• 23
• 22
• 21
• 20
• 19
• 18
• 17
• 16
• 15
• 14
• 13
• 12

Period in Months 12 6 4 3 2.4 2

#2 Diesel Monthplot, PPI, X11

Seasonal Means Seasonal Factors 0.92 0.93 0.94 0.95 0.96 0.97 0.98 0.99 1.00 1.01 1.02 1.03 1.04 1.05 1.06 1.07 1.08 1.09 1.10

Jan Feb Mar Apr May Jun Jul Aug Sep Oct Nov Dec

Series with weak seasonality (spectrum) Series X11 Q2 Model p(LB24) palm2 .56 011 .06 pbeef 1.59 011 .05 peqsw .66 011,011 .28 pfert .47 011,011 .18 pgasp .93 011 .03 plamb 1.14 011 .03 pplas .92 011 .77

Closing Remarks

• team approach working so far
• indications of improvement from model-

based for some series

• further investigation planned with sampling

error component

BLS Project Members Nicole Brooks, David Byun, Dan Chow, Tom Evans, Mike Giandrea, Raj Jain, Chris Manning, Jeff Medlar, Randall Powers, Stuart Scott, Eric Simants, Jeff Smith, Michael Sverchkov, Richard Tiller, Daniell Toth, Jeff Wilson Acknowledgments Agustin Maravall, David Findley, Brian Monsell, John Eltinge, Pat Getz