nrgi building an energy sentiment index peter deeney*, dublin city university business school...
TRANSCRIPT
NRGi
Building an Energy Sentiment Index
Peter Deeney*, Dublin City University Business School (DCUBS), IrelandMark Cummins, DCUBSAdam Bermingham,** DCU School of Computing Michael Dowling, DCUBS
*This material is based upon works supported by Dublin City University under the Daniel O'Hare Research Scholarship scheme.**CLARITY, Centre for Sensor Web Technologies, School of Computing, DCU, Dublin, Ireland. Work supported by Science Foundation Ireland under grant 07/CE/I1147
What’s new?
• Use of a Sentiment Index for Energy Market
• Use of Multiple Hypothesis Testing methods
What is Sentiment?• The propensity to speculate (Baker and Wurgler, 2006)
• Optimism or pessimism (Baker and Wurgler, 2006)
• A noisy signal which traders act upon (Brown, 1999)
• Non-informational trading (Tetlock, 2007)
• The gap between what we can explain and what we observe.
• The rational or irrational beliefs of market participants regarding price movements.
Hypothesis and NRGi Indices
• Changes in sentiment, measured by the NRGi indices, can account for part of the changes in fuel prices.
• Speculation Volume of Futures Trades, Paper:Physical
• Volatility VIX or V2X
• Fear Gold
• Optimism Volume of Futures in Arca Index
Principal Components• Brown and Cliff, 2004 and Baker and Wurgler, 2006• Subsequent components will not change the betas or the significance by much.• 1st Approximately the mean of the NRGi proxies with a sign change for Arca Index (XOI)
Brent Var Exp % 38 60 791 2 3
L Ln Brent Vol 0.51 0.10 -0.19L Ln Gold 0.51 0.03 -0.17P Ln XOI -0.25 0.59 -0.61P V2X 0.19 0.52 0.73P Spec 1 0.42 -0.46 -0.11L Spec 3 0.45 0.39 -0.12
Results for Ln Returns of BrentCoefficient Std. Error p – Value
Const -0.0023 0.0785 0.97619GDP Adj 0.11765 0.08056 0.14694OPEC Prop 0.0612 0.08213 0.45771Moody Aaa 0.22564 0.08289 0.00751 ***Brent PCA 1 -0.1269 0.05457 0.02184 **Brent PCA 2 0.30575 0.08269 0.00034 ***Brent PCA 3 -0.3841 0.08564 0.00002 ***
F (6,113) 8.079 R2 30%p F test 0.000000286 *** Adj. R2 26%Mean Dep Var -0.0066 Data Feb 2003 – Jan 2013
Std Dev Dep Var 1.00157 T = 120
Results for ModelsWTI Brent Australian Coal
Log Likelihood NRGi 0.0019 *** < 0.00005 *** 0.1404
Log Likelihood Surveys
0.0023 *** 0.00002 *** 0.6403 US0.253 EU
F test NRGi 0.0015 *** 0.0000008 *** 0.0027 ***
F test Surveys 0.00089 *** 0.00015 *** 0.052 * US0.051 * EU
Adj R2 NRGi 10% 21% 9.1%
Adj R2 Surveys 11% 14% 1.4% US4.0% EU
Brent and its NRGi (1)
-2.5
-2
-1.5
-1
-0.5
0
0.5
1
1.5
2
2004 2006 2008 2010 2012
LnWTIZBrentPca1
Brent and its Survey (1)
-3
-2.5
-2
-1.5
-1
-0.5
0
0.5
1
1.5
2
2004 2006 2008 2010 2012
LnWTIZEULevelPca1
Multiple Comparison Problem
Generalized Holm• Proposed by Lehmann and Romano, (2005)
• Lower value of p required to avoid Multiple Comparison Problem.
• k- FWER = P[reject at least k null hypotheses which are true]
• Significance level alpha set so that k-FWER < alpha, GH produces p value
• Alpha = 0.05 , k = 3 yields p = 0.0058, rejecting 28 hypotheses
• (Alpha = 0.04, k = 2 yields p = 0.0031, rejecting 28 hypotheses)
Thank You
• NRGi Energy Sentiment Index
• Multiple Hypothesis Testing
• Paper Available at www.peterdeeney.com