TRAMO/SEATS Quality Report Introduction This document presents the list of possible enhancement to the existing TRAMO/SEATS quality diagnostic in Demetra+. It is intended for discussion about possible improvements in diagnostic check for TRAMO/SEATS. The document is divided into two section. First one provides a current composition of Quality Report, displayed in Demetra+ “Main results” section. These measures determine the value of the final flag (bad/uncertain/good). Moreover, the list of additional quality measures displayed in the other part of the output is presented. Next part presents the list of indicators and statistics used in TRAMO/SEATS that can be add to the Quality Report and hence participate in a summary value. Also additional output, indicators and statistics used by original TRAMO/SEATS program that can be add to diagnostic section in Demetra+ have been listed. Those statistics are not expected to be added to the Quality Report, however as they are regarded to be highly informative, it is suggested to include them into Demetra+ diagnostics. 1. current situation Main quality measures included into Quality Report: No. 1 definition Name 2 annual totals 3 4 5 6 7 8 spectral seas peaks spectral td peaks normality independence spectral td peaks spectral seas peaks Description Inspection of the definition constraints (basic relationships between different components of the time series) The test compares the annual totals of the original series and those of the seasonally adjusted series. Visual significance of seasonal peaks Visual significance of trading day peaks The joint Doornik-Hansen normality test Ljung-Box test Test based on the periodogram of the residuals Test based on the periodogram of the residuals Section basic checks visual spectral analysis regarima residuals 1 No. 9 10 11 12 13 Name residual seasonality on sa residual seasonality on sa (last 3 years) residual seasonality on irregular number of outliers seas variance 14 irregular variance 15 seas/irr cross-correlation Description F-test on stable seasonality F-test on stable seasonality F-test on stable seasonality Number of outliers Variance of seasonal component, its theoretical estimator and empirical estimate Variance of irregular component, its theoretical estimator and empirical estimate Cross-correlation between seasonal and irregular component Section residual seasonality outliers seats Additional quality measures: No. Name Mean Skewness Kurtosis Normality Ljung-Box(16) (24 for monthly data) Box-Pierce(16) (24 for monthly data) Ljung-Box on seasonality(3) Box-Pierce on seasonality(3) Runs around the mean: number Runs around the mean: length Up and Down runs: number Up and Down runs: length Ljung-Box on squared residuals (16) (24 for monthly data) Box-Pierce on squared residuals (16) (24 for monthly data) Trend Description Distribution of the residuals Doornik-Hansen normality test Ljung-Box test on residuals Box-Pierce test on residuals Ljung-Box test on seasonal residuals Box-Pierce test on seasonal residuals Wald-Wolfowitz test Ljung-Box on squared residuals Section Pre-processing ->Residuals (Normality of the residuals) Pre-processing ->Residuals (Independence of the residuals) Pre-processing ->Residuals (Randomness of the residuals) Pre-processing ->Residuals (Linearity of the residuals) Box-Pierce on squared residuals Trend-cycle innovation variance Decomposition (Innovation 2 No. Name Seasonal Irregular trend sa seasonal irregular trend sa seasonal irregular trend/seasonal trend/irregular seasonal/irregular Friedman test Kruskall-Wallis test Test for the presence of seasonality assuming stability Evolutive seasonality test Combined seasonality test Residual seasonality test Relative (or absolute) difference in SA series Relative (or absolute) difference in trend Friedman test Kruskal-Wallis Evolutive seasonality test Indenifable seasonality test Abnormal values (%) Description Seasonal component innovation variance Irregular component innovation variance Comparison of variances among the theoretical component, theoretical estimator and empirical estimate – for each component (including transitory component if present) and for seasonally adjusted data Comparison of autocorrelation function of theoretical component, theoretical estimator and empirical estimate – for each component (including transitory component if present) and for seasonally adjusted data Comparison of cross-correlation between theoretical estimators with cross-correlation between empirical estimates Friedman test Kruskall-Wallis test Test for the presence of seasonality assuming stability Evolutive seasonality test Combined seasonality test Residual seasonality test Relative (or absolute) difference between the initial estimate and the latest estimate os SA series Relative (or absolute) difference between the initial estimate and the latest estimate of the trend-cycle Friedman test applied to the each span Kruskal-Wallis test applied to the each span Evolutive seasonality test applied to the each span Combined seasonality test applied to the each span % of ubnormal values in seasonal coponent, trading day component and seasonally adjusted series Section variance) Decomposition-> Model based tests Diagnostic ->seasonality tests Diagnostic ->Revision history Diagnostic ->Sliding spans 3 2. possible enhancement Indicators and statistics used in TRAMO/SEATS that can be add to the Quality Report: No. Indicator/statistics 1 Comparison of variances among the theoretical component, theoretical estimator and empirical estimate 2 3 4 5 6 Check of the cross-correlation between theoretical estimator and empirical estimate Check of autocorrelation coefficients of theoretical estimator and empirical estimate at lags 1 and 4 (or 12 for monthly series) Residual seasonality on SA, i.e. F-test on stable seasonality (on both the stochastic and the final SA component) Out of sample forecasting test Average of the absolute value of the differences between the annual averages of the original series, SA series and TC Available in Demetra+ Partially Yes Yes Partially No No Remarks It is already available in the QR for seasonal and irregular components (the cycle-trend component could be added) For seasonal-irregular is already available in the QR, trend-irregular and trend-seasonal could be added To be included in the QR This comparison may highlight the presence of outliers at the same month of different (contiguous) years Only the outcome of annual totals test is available now (I think that in the QR the result of the test is OK) Remark: Often the test on spectral seas peaks on residuals detects the presence of seasonality due to a peak on/near the last seasonal frequency. Perhaps the latter could be excluded from the current Quality report. 4 Other output, indicators and statistics used in TRAMO/SEATS that can be add to diagnostic section in Demetra+: No. 1 2 Standard deviation of the SEATS residuals Out of sample forecasting test in TRAMO part 3 Difference’s order in SEATS (when the model identified in TRAMO does not admit a decomposition) 4 5 Modulus and frequency of autoregressive roots Long term trend, the cycle and the stochastic trading day with forecasts 6 7 Innovation variance of component (not expressed in units of var(residuals)) Confidence intervals in the graphs (and tables) for the SA series, trend-cycle and business cycle1. 8 Period to period rate of growth (multiplicative model) or growth expressed in the series units (additive model) Annual rate of growth, centred at the last available observation and extended with forecast Standard errors of p-to-p rates of growth of trend and seasonally adjusted data Standard errors of rates of annual growth of trend and seasonally adjusted data % Reduction in revision S.E. after one year of data and five years 9 10 11 12 13 1 Indicator/statistics Standard error of both the total estimation error and the revision error in concurrent estimator (not expressed in units of var(residuals)) Remarks/source Suggested by A. Marvall to add it to Demetra+ Suggested by A. Marvall to add it to Demetra+ Now only frequency is available Suggested by A. Marvall to add it to Demetra+ Suggested by A. Marvall to add it to Demetra+ Only graphical representation is available now (panel Revision error in node Errors analysis) Only graphical representation is available now, Maravall A., Seasonal Adjustment and Signal Extraction in Economic Time Series Bussines cycle is not calculated in Demetra+ at the moment. 5