Presentation is loading. Please wait.

Presentation is loading. Please wait.

Overview of Main Quality Diagnostics Anu Peltola Economic Statistics Section, UNECE UNECE Workshop on Seasonal Adjustment 20 – 23 February 2012, Ankara,

Similar presentations


Presentation on theme: "Overview of Main Quality Diagnostics Anu Peltola Economic Statistics Section, UNECE UNECE Workshop on Seasonal Adjustment 20 – 23 February 2012, Ankara,"— Presentation transcript:

1 Overview of Main Quality Diagnostics Anu Peltola Economic Statistics Section, UNECE UNECE Workshop on Seasonal Adjustment 20 – 23 February 2012, Ankara, Turkey

2 UNECE Statistical Division Slide 2February 2012 Overview  Purpose of quality diagnostics  Main quality issues  Main results  First visual checks  Pre-processing  Decomposition  Main quality diagnostics

3 UNECE Statistical Division Slide 3February 2012 Purpose of Quality Diagnostics  Seasonality is identified based on hypotheses Seasonal component is estimated = is uncertain  Diagnostics will reveal any essential weaknesses in seasonal adjustment  Help draw attention to problematic issues They prevent the use of misleading results that could lead to false signals  The automatic procedure in Demetra+ is reliable! But diagnostics are especially important for analysing in detail the aggregate series

4 UNECE Statistical Division Slide 4February 2012 Main Quality Issues  Appropriateness of the identified model and components  Number and type of outliers  Stability of the seasonal component  Absence of residual seasonality and residual calendar effects  Magnitude of the possible phase delay

5 UNECE Statistical Division Slide 5February 2012 Main results inform you about…  Estimation time span used for identifying the seasonal pattern  Application of log-transformation  If there working day, Easter or Leap year effects were identified  If outliers were found and when  A summary quality diagnostics

6 UNECE Statistical Division Slide 6February 2012 Visual checks  To find seasonal breaks and high variability  Problematic with moving averages, fitting the ARIMA model and finding effects

7 UNECE Statistical Division Slide 7February 2012 Pre-processing  Statistical properties of the ARIMA model  Regression variables  The pre-adjusted series  Residuals should be independent and random and follow normal distribution

8 UNECE Statistical Division Slide 8February 2012 Decomposition  Stochastic series presents the results  Cross-correlation of results In theory, components should be uncorrelated A green p-value in Demetra+ would indicate insignificant cross-correlation EstimatorEstimateP-Value Trend/Seasonal-0.1250-0.15040.8018 Trend/Irregular-0.0450-0.08560.7311 Seasonal/Irregular 0.0446 0.01950.5900

9 UNECE Statistical Division Slide 9February 2012 Quality Diagnostics  Presence of seasonality  Spectral graphics  Revision history  Sliding spans  Model stability analysis

10 UNECE Statistical Division Slide 10February 2012 Presence of Seasonality  Friedman test & Kruskall-Wallis test Is there stable seasonality?  Evolutive seasonality test Is there moving seasonality?  Combined seasonality test Is there identifiable seasonality?  Residual seasonality test Is there seasonality left in residuals in the entire series or in the last 3 years of data?

11 UNECE Statistical Division Slide 11February 2012 Spectral Graphics  Periodogram  Auto-regressive spectrum Analyse the residuals, irregular component and seasonally adjusted series for remaining seasonal or trading day effects Spectral graphics of the residuals

12 UNECE Statistical Division Slide 12February 2012 Revision History  Analyses revisions that happen when new observations are added at the end of the series

13 UNECE Statistical Division Slide 13February 2012 Sliding Spans  Analyses stability of Seasonal component Trading day effect (if present) Seasonally adjusted series Slidings spans of the seasonal component

14 UNECE Statistical Division Slide 14February 2012 Model Stability Analysis  Calculates ARIMA parameters and coefficients of regression variables for different periods  Computes the results on a moving window of eight years which slides by one year  The points correspond to the successive estimations  Strong movement of values from negative to positive indicates instability

15 UNECE Statistical Division Slide 15February 2012 Problematic Issues  Which are the most essential tests?  How to read and understand the diagnostics?  When does a result signify bad quality?  What to do to improve results?  Which poor results of quality diagnostics could be accepted?  Which quality diagnostics could be published to the users?


Download ppt "Overview of Main Quality Diagnostics Anu Peltola Economic Statistics Section, UNECE UNECE Workshop on Seasonal Adjustment 20 – 23 February 2012, Ankara,"

Similar presentations


Ads by Google