1 / 51

Real-Time Signalextraction

This research focuses on developing a customized estimation criteria for real-time signal extraction. The application of this criteria is demonstrated using the KOF-economic barometer for GDP estimation. The aim is to reduce delay and improve reliability of real-time estimates simultaneously.

ldye
Download Presentation

Real-Time Signalextraction

An Image/Link below is provided (as is) to download presentation Download Policy: Content on the Website is provided to you AS IS for your information and personal use and may not be sold / licensed / shared on other websites without getting consent from its author. Content is provided to you AS IS for your information and personal use only. Download presentation by click this link. While downloading, if for some reason you are not able to download a presentation, the publisher may have deleted the file from their server. During download, if you can't get a presentation, the file might be deleted by the publisher.

E N D

Presentation Transcript


  1. Real-Time Signalextraction An Application of Customized Estimation Criteria

  2. Real-Time Estimation:KOF-economic barometer

  3. GDP and old KOF-economic barometer (ex post) Standardized growth rates 3 3 2 2 GDP 1 1 0 0 -1 -1 -2 -2 KOF-Barometer -3 -3 84 86 88 90 92 94 96 98 00 02 04

  4. `Old´ KOF-economic barometer (real time) 1.5 0 -1.5 2001 2002 2003

  5. Problems • Real-time estimates lag • Delay • Real-time estimates are noisy • Unreliability • Purpose of TP-criterion • Reduce delay and improve reliability, SIMULTANEOUSLY!

  6. Signal Definition Customization

  7. Definition • Signal: match intention of researcher/user • KOF economic barometer: • Eliminate short-term (periodicity < 1.5 years) fluctuations from a leading indicator of GDP • Transfer function of the filter • Symmetric filter

  8. Business-cycle frequencies,cutoff: 1.5 years

  9. Symmetric MA-filter

  10. Customization • Specify a smoothing process that fits your intention • Signal • Estimate that signal properly in real-time • Concurrent filter (completely asymmetric) • Customized estimation criteria

  11. Research Focus • Business cycle analysis • Trading, climatology, linguistics • Detection of turning-points • Local extrema of the (previous) trend component • Experience: these TP‘s are informative about the future dynamics of the series • Convey a strong prospective content

  12. Real-time estimation problem Traditional MBA (level) Methodological issues

  13. Traditional MBA • X-12, X-11 ARIMA, TRAMO/SEATS • Identify a time series model for the DGP • Forecast the future • Apply the symmetric filter to the extended time series • Criterion • One-step ahead mean-square performance • Multi-step ahead forecasts needed • Mismatch because models are misspecified

  14. TRAMO: heavy misspecification(120 Obs.) • Business survey data: bounded time series

  15. Conclusions • Statistics based on one-step ahead forecasts • Estimation, Identification (AIC/BIC, unit-roots tests,…), Diagnostics (Box-Pierce, Ljung-Box) • Cannot detect misspecification related to mid-or long-term dynamics • Conflict one vs. multi-step ahead forecasting performances • Consequences • Inefficient real-time filters • Unnecessarily large time delays • TP cannot be accounted for explicitly

  16. Customized criteria Level Turning Point (TP)

  17. Material • Paper • Book • Marc.wildi@zhawin.ch • Software • R-package: signalextraction • CRAN (ETHZ, Zurich)

  18. Empirical results TP‘s (Part 1) Simple level criteria (Benchmarks)

  19. Criteria (Level) • Customization: WYOIWYG

  20. Simple designs • Optimal (mean-square) real-time level filter • L-L filter • TP: local extrema • Optimal (mean-square) real-time level filter in first differences • L-D Filter • TP: crossings of the zero-level line

  21. L-L and L-D filters

  22. TP‘s: identification in real-time

  23. Conclusions • Level-Filter for original data (L-L-Filter) • Faster • Noisier • Classical trade-off • Next steps: improve • Speed • Speed AND reliability

  24. Empirical results TP‘s (Part 2) Improving speed An explicit focus towards TP‘s

  25. Characteristics of previous level filters

  26. Controlling the time delay • λ>1: emphasize the time delay in the pass-band • λ=1: best level filter • Issues are detailed in paper/book

  27. Empirical results part 2 • Two competing designs • Optimal real-time level filters in first differences • λ=1: mean-level performance • λ=6: SELECTIVE mean-level performance

  28. L-L (black) vs. L-D (λ=1: red and λ=6: brown) • Rate of false alarms: 9.1%(λ=6), 12.4% (λ=1),15% (level)

  29. Filter characteristics

  30. Selectivity: improved performance in TP‘s

  31. Selectivity: improved performance in TP‘s • Mean-square filter errors • Selectivity-effect • Obtained without knowledge of location of TP‘s • Due to stylized fact: second derivative large (in absolute value) in TP‘s

  32. Empirical results TP‘s (Part 3) Improving speed and reliabilitySIMULTANEOUSLY Research Projects

  33. Customization: Controlling time delay and high-frequency damping • Stronger damping of high-frequency noise in stop-band • Smaller time delays in pass-band • W(ω) is monotonic (increasing) and λ>1

  34. Competing Designs • Fast real-time level filter in first differences • λ=6 • Logit-model • New TP-filter

  35. Comparison: TP-filter vs fast (λ=6) level-filter

  36. TP-filter vs. logit-model

  37. Performances • Rate of undesirable errors: • TP: 4.1% (IN REAL-TIME) • Logit: 9.1% • λ=6: 9.1%

  38. Amplitude DFA TP-filter vs. DFA level filter (KOF-Economic Barometer)

  39. Delay TP-filter vs. level filter(KOF-Economic Barometer)

  40. Conclusions • TP-filter is both fast and reliable • There is a trade-off! • But optimal filter designs can significantly improve upon • Traditional level filters • Logit models • Traditional level- and TP-criteria are incongruent • Generalization of level-criterion • Improved level performance in TP‘s

  41. Economic Sentiment Indicator (ESI) for the euro area, published monthly by DG ECFIN. DFA vs Dainties

  42. In the past (DFA: fast and reliable)

  43. Towards the current boundary

  44. Leading Indicator for German GDP Research Project with the Bundesministerium für Wirtschaft und Technologie

  45. GDP and fastest (real-time) series

  46. Comparison: symmetric trend and real-time output

  47. Conclusions

  48. Optimization criteria • Very often ML does not fit the intention of the researcher/user • Model misspecification: functions of ML-estimates are not ML-estimates of these functions • Example: multi-step ahead forecasts • Customization • Define and implement criteria that match the intention of the researcher • Very often: generalization of ML

  49. Outcome • New estimation criteria • Level/TP’s • Estimate parameters of real-time filter directly • New diagnostics • Verify characteristics of real-time filter directly • New filter constraint test • Generalizes unit root tests to problems involving one- and multi-step ahead forecasts

  50. New Results/Work in Progress • Trading filters • Non-linear Filters • Asymmetric cycles • Multivariate filters

More Related