NDM Data Sample Analysis: Final Results (2)
description
Transcript of NDM Data Sample Analysis: Final Results (2)
© British Gas Trading Limited 2011
NDM Data Sample Analysis: Final Results (2)
Option C: Regression Analysis
5th December 2012
© British Gas Trading Limited 2011
Contents
• In-Sample Results as per previous presentation i.e., Winter/Summer Modelling on 2008/2009 Data
• AQ-Corrected Out-of-Sample Model fit using 2009/2010 data
• Conclusion
© British Gas Trading Limited 2011
2009/2010 GAS YEAROUT-OF-SAMPLE RESULTS
© British Gas Trading Limited 2011
Out-of-Sample Results: EUC1 Graphical Output
EUC1 Oct 09-Mar 10 XoServe AQ NDM Sample AQMAPE 7.68% 4,134,029 3,803,042 Average BIAS 7.12%R² 97.87%
© British Gas Trading Limited 2011
Out-of-Sample Results: EUC2 Graphical Output
EUC2 Oct 09-Mar 10 XoServe AQ NDM Sample AQMAPE 16.23% 16,592,319 5,350,971 Average BIAS 9.62%R² 88.26%
© British Gas Trading Limited 2011
Out-of-Sample Results: EUC3 Graphical Output
EUC3 Oct 09-Mar 10 XoServe AQ NDM Sample AQMAPE 21.89% 45,173,231 20,269,895 Average BIAS 19.87%R² 93.89%
© British Gas Trading Limited 2011
Out-of-Sample Results: EUC4 Graphical Output
EUC4 Oct 09-Mar 10 XoServe AQ NDM Sample AQMAPE 13.67% 440,453,374 214,592,141 Average BIAS 12.12%R² 93.81%
© British Gas Trading Limited 2011
Conclusion
• Out-of-Sample data fit for Oct-09 to Mar-10 only as significant level shift (i.e., 30%+) observed. This is could be because of Regression coefficients needing to be re-estimated.
• Out-of-sample accuracy results poor. This could be because of
a) the number of meters changing year on year
b) Lack of normalisation by AQ prior to modelling
c) Significant behavioural change not captured by analysis
d) AQ re-base issue