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

relative contribution of components to change in the observed series n X 1 ∑ = − t X 1 mult. case − d n d X = + − t d 1 t d = = X O T S I , , , and d 1 ,3,12,24 = + + '2 2 2 2 O T S I d d d d rel. contribution × 2 '2 100% S / O d d of 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

Meats Autoregressive Spectrum (Decibels), PPI, SEATS Unadjusted Seasonally Adjusted Visually Significant Peak Visually Significant Peak Unadjusted Median Seasonally Adjusted Median -25 TD TD TD -26 -27 -28 -29 -30 -31 -32 -33 -34 -35 -36 -37 -38 -39 -40 -41 12 6 4 3 2.4 2 Period in Months

Meats Monthplot, PPI, SEATS Seasonal Means Seasonal Factors 1.03 1.02 1.01 1.00 0.99 0.98 0.97 Apr May Aug Sep Jan Mar Jun Nov Dec Feb Jul Oct

Meats Monthplot, PPI, X11 Seasonal Means Seasonal Factors 1.04 1.03 1.02 1.01 1.00 0.99 0.98 0.97 Apr May Aug Sep Jan Mar Jun Nov Dec Feb Jul Oct

Publications Autoregressive Spectrum (Decibels), PPI, SEATS Unadjusted Seasonally Adjusted Visually Significant Peak Visually Significant Peak Unadjusted Median Seasonally Adjusted Median 10 TD TD TD 0 -10 -20 12 6 4 3 2.4 2 Period in Months

Publications Autoregressive Spectrum (Decibels), PPI, X11 Unadjusted Seasonally Adjusted Visually Significant Peak Visually Significant Peak Unadjusted Median Seasonally Adjusted Median 10 TD TD TD 0 -10 -20 -30 12 6 4 3 2.4 2 Period in Months

Publications Monthplot, PPI, X11 Seasonal Means Seasonal Factors 0.9 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 Apr May Aug Sep Jan Mar Jun Nov Dec Feb Jul Oct

Scenic Trans, CES, SEATS Series scentran NBER Recessions in Gray Unadjusted Seasonally Adjusted Trend 36000 36000 35000 35000 34000 34000 33000 33000 32000 32000 31000 31000 30000 30000 29000 29000 28000 28000 27000 27000 26000 26000 25000 25000 24000 24000 23000 23000 22000 22000 21000 21000 20000 20000 19000 19000 1-95 1-96 1-97 1-98 1-99 1-00 1-01 1-02 1-03 1-04 1-05

Scenic Trans Monthplot, CES, SEATS Seasonal Means Seasonal Factors 1.4 1.3 1.2 1.1 1.0 0.9 0.8 0.7 Apr May Aug Sep Jan Mar Jun Nov Dec Feb Jul Oct

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

#2 Diesel Autoregressive Spectrum (Decibels), PPI, X11 Unadjusted Seasonally Adjusted Visually Significant Peak Visually Significant Peak Unadjusted Median Seasonally Adjusted Median -12 TD TD TD -13 -14 -15 -16 -17 -18 -19 -20 -21 -22 -23 -24 -25 -26 -27 -28 -29 -30 -31 12 6 4 3 2.4 2 Period in Months

#2 Diesel Monthplot, PPI, X11 Seasonal Means Seasonal Factors 1.10 1.09 1.08 1.07 1.06 1.05 1.04 1.03 1.02 1.01 1.00 0.99 0.98 0.97 0.96 0.95 0.94 0.93 0.92 Apr May Aug Sep Jan Mar Jun Nov Dec Feb Jul Oct

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