What Drove the Mid-2000s Explosiveness in Alternative Energy Stock Prices?

CIGI PAPERS

NO. 36 — JULY 2014

WHAT DROVE THE MID-2000s’ EXPLOSIVENESS IN ALTERNATIVE ENERGY STOCK PRICES? EVIDENCE FROM US, EUROPEAN AND GLOBAL INDICES

MARTIN T. BOHL, PHILIPP KAUFMANN AND PIERRE L. SIKLOS

WHAT DROVE THE MID-2000s’ EXPLOSIVENESS IN ALTERNATIVE ENERGY STOCK PRICES? EVIDENCE FROM US, EUROPEAN AND GLOBAL INDICES Martin T. Bohl, Philipp Kaufmann and Pierre L. Siklos

Copyright © 2014 by the Centre for International Governance Innovation The opinions expressed in this publication are those of the authors and do not necessarily reflect the views of the Centre for International Governance Innovation or its Operating Board of Directors or International Board of Governors or the United Nations University.

This work is licensed under a Creative Commons Attribution — Non-commercial — No Derivatives License. To view this license, visit (www.creativecommons.org/ licenses/by-nc-nd/3.0/). For re-use or distribution, please include this copyright notice.

ACKNOWLEDGEMENTS Part of this research was conducted while Philipp Kaufmann was visiting the Viessmann European Research Centre at Wilfrid Laurier University. The authors are grateful to the Centre for International Governance Innovation for financial assistance through a Collaborative Research Grant.

57 Erb Street West Waterloo, Ontario N2L 6C2 Canada tel +1 519 885 2444 fax +1 519 885 5450 www.cigionline.org

TABLE OF CONTENTS 4

About the Authors

4

About the Project

4 Acronyms 4

Executive Summary

4 Introduction 5 Methodology 5

Performance Measurement Models

6

Bubble Detection Tests

7

Construction of Idiosyncratic Price Time Series

8 Data 9

Empirical Results 9

Performance Analysis

10

Idiosyncratic Price Time Series and Factor Exposures

13

SADF tests

20 Works Cited 22 About CIGI 22 CIGI Masthead

CIGI Papers no. 36 — July 2014

ACRONYMS ABOUT THE AUTHORS Martin T. Bohl is professor of economics, Centre for Quantitative Economics, Westphalian Wilhelminian University of Münster. From 1999 to 2006, he was a professor of finance and capital markets at the European University Viadrina Frankfurt (Oder). His research focuses on monetary theory and policy as well as financial market research.

Philipp Kaufmann is a research associate, doctoral candidate and the Chair of Monetary Economics at the Westfälische Wilhelms-University Münster.

Prior to this, he was a visiting researcher at Wilfrid Laurier University in Waterloo, Canada. He completed a double degree program at the Ecole de Management Strasbourg at the University of Strasbourge, majoring in finance, and received a diploma in business administration at the Friedrich-Alexander-University ErlangenNuremberg. His research focuses include price and earnings momentum, mutual fund performance, investor sentiment and alternative energy stocks.

Pierre L. Siklos is a CIGI senior fellow and a research associate at Australian National University’s Centre for Macroeconomic Analysis.

His research interests are in applied time series analysis and monetary policy, with a focus on inflation and financial markets.

ABOUT THE PROJECT This publication emerges from a project called Essays in Financial Governance: Promoting Cooperation in Financial Regulation and Policies.

The project is supported by a 2011-2012 CIGI Collaborative Research Award held by Martin T. Bohl, Badye Essid, Arne Christian Klein, Pierre L. Siklos and Patrick Stephan. In this project, researchers investigate empirically policy makers’ reactions to an unfolding financial crisis and the negative externalities that emerge in the form of poorly functioning financial markets. At the macro level, the project investigates whether the bond and equity markets in the throes of a financial crisis can be linked to overall economic performance. Ultimately, the aim is to propose policy responses leading to improved financial governance.

ADF

Augmented Dickey-Fuller

BSADF backward supremum augmented Dickey-Fuller CPI

consumer price indices

ERIX

European Renewable Energy Index

GSADF generalized supremum Augmented DickeyFuller GSCI

Goldman Sachs Commodity Index

IMI

Investable Market Indices

MSCI

Morgan Stanley Commodity Index

NYSE

New York Stock Exchange

REN21 Renewable Energy Policy Network for the 21st Century SADF

supremum Augmented Dickey-Fuller

UNDP United Nations Development Programme WTI

West Texas Intermediate crude oil

EXECUTIVE SUMMARY Soaring prices in European alternative energy stocks and their subsequent tumble have attracted attention from both investors and academics. This paper extends recent research to an international setting and analyzes whether the explosive price behaviour of the mid-2000s was driven by rising crude oil prices and an overall bullish market sentiment. Inflation-adjusted US alternative energy stock prices do not exhibit signs of explosiveness. By contrast, we find strong evidence of explosive price behaviour for European and global sector indices, even after controlling for a set of explanatory variables. Interestingly, while the sector indices plunged with the outbreak of the global financial crisis, idiosyncratic components continued to rise and did not start to decline until after world equity markets had already begun to recover in 2009. This finding suggests a substantial revaluation of alternative energy stock prices in light of intensifying sector competition and shrinking sales margins, and casts some doubts on the existence of a speculative bubble. Nevertheless, this paper observes temporary episodes of explosiveness between 2005 and 2007 followed by rapid collapses, indicating the presence of some irrational exuberance among investors.

INTRODUCTION Optimistic investor sentiment is frequently viewed as having triggered the exceptional performance of renewable energy stocks in the mid-2000s. Promising

4 • CENTRE FOR INTERNATIONAL GOVERNANCE INNOVATION

WHAT DROVE THE MID-2000s’ EXPLOSIVENESS IN ALTERNATIVE ENERGY STOCK PRICES? EVIDENCE FROM US, EUROPEAN AND GLOBAL INDICES growth prospects and the anticipation of intensified government support even led to spikes in the stock price indices. The rapid growth and deployment of renewable energy sources has been promoted across the globe.

According to Bloomberg New Energy Finance (2013), the global new investment volume in renewable energy rose from US$40 billion in 2004 to US$244 billion in 2012. Using a multi-country, fixed-effects panel approach, Eyraud, Clements and Wane (2013) find that green investment has been stimulated worldwide by economic growth, low long-term nominal interest rates, high fuel prices and the adoption of certain policy instruments, such as feed-in tariffs or carbon-pricing schemes. These types of support policies have been put into practice by 127 countries (see the Renewable Energy Policy Network for the 21st Century [REN21] 2013), whereas 138 countries have even set specific policy targets aimed at increasing the share of renewables in both electricity production and final energy consumption. Based on 2011 data assembled

by REN21, the estimated renewable energy share of global energy consumption amounts to 9.7 percent, which is well short of most countries’ long-term goals (ibid.). While new investment volumes are still higher in developed countries, emerging economies have experienced more stable growth paths and have been catching up recently (Bloomberg New Energy Finance 2013). In 2012, the three leading regions with respect to new investment activities were Europe, China and the United States.

Despite the promising future for renewable energy technologies, fierce competition and excess supply from Asian manufacturers have taken their toll on the sector’s profit margins since the late 2000s. As a result, investor sentiment toward these fad stocks began to gradually deteriorate. Following the outbreak of the global financial and economic crisis, prices of alternative energy stocks plunged as quickly as they had risen, resulting in an almost hump-shaped performance pattern. Visual inspection of the price charts appears to suggest a speculative price bubble prior to 2008. However, previous research has not yet investigated whether there are exogenous risk factors that may have driven this apparent explosive behaviour.

Intuitively, soaring crude oil prices could have promoted exuberance in renewable energy stocks. Furthermore, many have found price movements in technology stocks to be strongly correlated with those of alternative energy stocks (see Henriques and Sadorsky 2008; Kumar, Managi and Matsuda 2012; Sadorsky 2012a). Given the high market betas that renewable energy stocks usually possess, it also seems possible that the pronounced bull market between 2003 and 2007 partly encouraged the bubble-like behaviour.

This paper aims to remove the systematic component of stock price movements and to focus the analysis on the idiosyncratic part of alternative energy stock prices.

For US alternative energy stocks, there is no evidence of an explosive root. By contrast, for European and global

stock indices, even inflation-adjusted idiosyncratic price time series exhibit explosive behaviour. This paper concludes that the bubble-like pattern seen in the data was mostly sector-specific and cannot entirely be attributed to exogenous risk factors. Surprisingly, the idiosyncratic price components did not plummet simultaneously with the price indices during the global financial crisis, but instead continued to increase until May 2009. This finding suggests that the price crash in 2008 was not the burst of a bubble, but rather the result of the stocks’ elevated sensitivity to market fluctuations. Nevertheless, this paper argues that corrections to price spikes observed between 2005 and 2007 indicate the presence of some sentimentdriven investor overreaction. Interestingly, the assumed positive correlation with crude oil prices only holds until the bursting of the oil price bubble in the second half of 2008. In the period that followed, the relationship breaks down and weakens considerably.

Furthermore, we find that our US, European and global sector indices are tilted toward the small-cap and growth stock segment. A subperiod analysis also reveals that alternative energy stocks belonged to the group of winner stocks between 2004 and 2007. Winner stocks are defined as those stocks in the cross-section of the market that performed relatively well over the previous year and tend to continue their outperformance in subsequent months — a phenomenon widely referred to as momentum. However, in the course of the global financial crisis, alternative energy stock indices have lost their positive momentum and have even produced significantly negative multifactor abnormal returns.

METHODOLOGY

PERFORMANCE in the second half of 2008. MEASUREMENT In the period that followed, the relationship breaks down and w considerably.

MODELS

Furthermore, we find that our US, European and global sector indices are tilted toward th Previous research suggests that the performance patterns and growth stock segment. A subperiod analysis also reveals that alternative energy stock of alternative energy stocks differ substantially from to the group of winner stocks between 2004 and 2007. Winner stocks are defined as those thosethe ofcross-­‐section conventional large-cap stocks.

While the of the market that performed relatively well o ver tresults he previous year and in Henriques and Sadorsky (2008) as well as Sadorsky continue their outperformance in subsequent months — a phenomenon widely referred t momentum.

However, the crenewable ourse of the global financial stocks crisis, alternative (2012b) indicate thatin US energy belongenergy stock lost high-beta their positive msegment, omentum and have even produced significantly negative multifactor a to the Bohl, Kaufmann and Stephen returns.

(2013) report a pronounced small-cap tilt and timevarying momentum exposure for their German sample.

To gain insight into how systematic factors contribute to Methodology the performance of our international sector indices, we Performance Measurement Models employ the Carhart (1997) four-factor model:

Previous research suggests that the performance patterns of alternative energy stocks diff substantially from those of conventional large-­‐cap stocks. While the results in Henriques a (2008) as well as Sadorsky (2012b) indicate that US renewable energy stocks belong to the (1) segment, Bohl, Kaufmann and Stephen (2013) report a pronounced small-­‐cap tilt and time momentum exposure for their German sample. To gain insight into how systematic factor of our international sector indices, we erisk-free mploy the Carhart –to rftthe isperformance the index excess return over the rate (1997) four model: month t. The unconditional Carhart (1997) four-factor

Rt in alphaR а−4Fr represents the+ βabnormal return after adjusting 4F + β1RMRF t ft = α t 2 SMBt + β3 HMLt + β 4WMLt + ε t . for sensitivities to the four systematic risk factors.

RMRF

Rt – rft is the index excess return over the risk-­‐free rate in month t. The unconditional Carh

four-­‐factor alpha α 4 F represents the abnormal return after adjusting for sensitivities to t systematic risk fT. actors.

RMRF denotes the value-­‐weighted market portfolio Martin Bohl, Philipp Kaufmann and Pierre L. Siklos •r5eturn in exces free rate. The return difference between small-­‐cap and large-­‐cap stocks is captured by SM HML measures the return spread between high and low book-­‐to-­‐market equity stocks. WM for the price momentum factor defined as the difference between the returns of past win

financial crisis. To allow for such a potential change in the parameters, we additio

to the group of winner stocks between 2004 and 2007. Winner stocks are defined as those stocks in variable regression: the cross-­‐section of the market that performed relatively well over the previous year and tend to continue their outperformance in subsequent months — a phenomenon widely referred to as momentum. However, in the course of the global financial crisis, alternative energy stock indices have Rt − rft = α15 F + α 25 F Dt lost their positive momentum and have even produced significantly negative multifactor abnormal returns.

CIGI Papers no. 36 — July 2014

5

+ ∑ ( β1k + β2 k Dt ) Fkt + ε t , k =1

denotes the value-weighted market portfolio return in tests (Phillips, Wu and Yu 2011) as well as the generalized where Fkt denotes one of the five factor portfolios of equation (2). The dummy v excess of the risk-free rate.

The return difference between SADF (GSADF) version introduced by Phillips, Shi and Methodology to zero from January 2004 to Dunit ecember and equal proved to one from January 2008 small-cap and large-cap stocks is captured by SMB, while Yu (2013).

These right-tailed root2007 tests have Performance Measurement odels between high and low dummy coding enables us to examine the abnormal behaviour performance and factor expo HML measures the returnMspread useful in detecting exuberance or bubble-like Previous research suggests that the performance patterns of alternative energy stocks differ in financial after the otime utbreak of the global It is also consistent book-to-market equity stocks.

WML stands for the price series and arefinancial appliedcrisis. to daily real price with the fact th substantially from those of conventional large-­‐cap stocks. While the results in Henriques and Sadorsky 2 2 momentum factor defined as the difference between the data.

The SADF tests are based on the assumption that renewable e nergy s tock i ndices p eaked a t t he e nd o f 2 007.

(2008) as well as Sadorsky (2012b) indicate that US renewable energy stocks belong to the high-­‐beta returns of past winner and loser stocks.

The error term is asset prices follow a random walk and thus contain a unit segment, Bohl, Kaufmann and Stephen (2013) report a pronounced small-­‐cap tilt and time-­‐varying Bubble Detection Tests denoted ԑt. for their German sample. To gain insight into how systematic factors croot.

Exceptions to the rule are strong upward departures momentum by exposure ontribute to the performance of our international sector indices, we employ the Carhart (1997) four-­‐factor from values, which can lead to explosiveness To fundamental uncover potential explosiveness in the deflated price time series of renewable Henriques and Sadorsky (2008), Kumar, Managi and model: in the time series. The recursive SADF (SADF) tests (Ph we uunderlying se recursive aprice nd rolling supremum Augmented Dickey-­‐Fuller Matsuda 4(2012), as well as Broadstock, Cao and Zhang test estimates the conventional Augmented Dickey-Fuller F Rt − rft = α + β1RMRFt + β2 SMBt + β3 HMLt + β4WMLt + ε t .

(1) 2011) as well as the generalized SADF (GSADF) version introduced by Phillips, Shi

(2012) document a significant influence of crude oil prices (ADF) regression bypusing forward expanding right-­‐tailed unit repeatedly root tests have roved uaseful in detecting exuberance or bubble-­‐ on alternative Sadorsky (2012a)Carhart sample Rt –the rft is prices the index of excess return over tenergy he risk-­‐free stocks. rate in month t. The unconditional (1997) sequence: 3 financial t ime s eries a nd a re a pplied t o d aily r eal p rice d ata.

The SADF tests are b 4F even suggests athe short position in acrude futures to the four represents abnormal return after djusting oil for sensitivities four-­‐factor alpha α entering that asset prices follow a random walk and thus contain a unit root. E factors. RMRF denotes clean the value-­‐weighted arket portfolio return in excess of the assumption risk-­‐ tosystematic hedgerisk against falling energy mstock prices.

We free rate. The extend return difference etween small-­‐cap model and large-­‐cap is captured by SMB, while are strong upward departures from fundamental values, which (4) can lead to explos therefore the bfour-factor bystocks additionally HML measures the return spread between high and low book-­‐to-­‐market equity stocks. WML stands underlying price time series. The recursive SADF test estimates the conventional A controlling for return variations in futures contracts of for the price momentum factor defined as the difference between the returns of past winner and loser Fuller regression repeatedly by using a forward expanding fossil fuels. This results in the following specification: where y (ADF) is the daily real alternative energy stock price sample sequence stocks. The error term is denoted by ε . t

t

index, ∆ stands for the P first difference operator, and μ, δ Henriques and Sadorsky (2008), Kumar, Managi and Matsuda (2012), as well as Broadstock, Cao and are regression The error term ԑt is and ϕ Δy = µ + δyt −1 + ∑φcoefficients. p Δyt − p + ε t , Zhang (2012) document a significant influence of crude oil prices on the prices o(2) f alternative energy p t p =1 independent and identically distributed with zero mean

stocks. Sadorsky (2012a) even suggests entering a short position in crude oil futures to hedge against and constant variance.

To determine the optimal lag length falling clean energy stock prices. We therefore extend the four-­‐factor model by additionally controlling where а5F denotes the monthly five-factor abnormal return real alternative stock rice index, Δ stands for the f yt is the daily regression, for return variations in futures contracts of fossil fuels. This results in the following specification: P inwhere each subsample weenergy follow thepprocedure and Energy is a proxy for the excess return on investments suggested by Campbell and Perron (1991). Starting with = α 5 F + β1commodity RMRFt + β2 SMBtmarket. + β3 HMLt + β4WMLt + β5 Energyt + ε t , (2) operator, and µ , δ and φ p are regression coefficients. The error term ε t is inde t − rftenergy inRthe

six lags, we reduce the lag order until the coefficient on the identically distributed with zero mean and constant variance. To determine the o 5F where α denotes the monthly five-­‐factor abnormal return and Energy is a proxy for the elast xcess included lag is significant at the five percent level (see Bohl, Kaufmann and Stephan (2013) also uncover some subsample regression, we follow the procedure suggested by Campbell and alsoeach Phillips, Wu and Yu 2011). return on investments in the energy commodity market. considerable time variation in price momentum exposures Starting with six lags, we reduce the lag order until the coefficient on the last inclu Bohl, risk-adjusted Kaufmann and Stephan (2013) also uncover some considerable time variation of in price and returns, especially after the outbreak A right-tailed hypothesis testPhillips, is conducted on the at the five percent level (see also Wu and Yu 2011). momentum exposures and risk-­‐adjusted returns, especially fter the ofor utbreak of the a2008-­‐2009 global the 2008-2009 global financial crisis.

To aallow such supremum test statistic, which is determined by the potential change in the parameters, we additionally run a A right-­‐tailed hypothesis test is conducted osequence n the supremum test statistic, which maximum value of the corresponding of ADF 3 dummy variable regression: statistics δ. vFollowing Phillips, Wu and Yu we δ . Following Ph maximum alue of the corresponding sequence of A(2011), DF statistics determine initial window length theby integer part (2011), wthe e determine the initial w indow lby ength the integer part of Tr0 , where of Tr0, where T denotes the total sample size and the (3) fractionrun r is equal to 0.10. Given our sample size of 2,406 financial crisis. To allow for such a potential change in the parameters, we additionally 0 a d ummy

trading days, the initial window roughly covers the first 2 variable regression: We also use the Quandt-Andrews unknown breakpoint test (Andrews 1993; Andrews where Fkt denotes one of the five factor portfolios of sample yearcheck andwhether therefore a sufficient number of formally there yields is a structural change in all of the five-factor model’s para 5 equation (2). The variable Dt is equal to zero from observations to ensure estimation efficiency.

The window 5 F dummy 5F that a breakpoint occurs in February 2008 for the US stock index, in January 2008 for th R − rft = α1 + α 2 Dt + ∑ ( β1k + β2 k Dt ) Fkt + ε t , (3) January t 2004 to December 2007 and equal to one from size expands observation after each pass.

Hence, k =1 and in July by 2008one for the global stock indices.

For ease of comparability, we date only o January 2008 to July 2013. This dummy coding enables us the recursive statistic is defined as: in January SADF 2008 in the regression specifications. where the enotes one operformance f the five factor portfolios of equation (2). The dummy variable Dt is equal Fkt dabnormal to examine and factor exposures 3 SADF-type tests have been employed to test for speculative bubbles in equity (Phil to and zero fafter rom January 2004 to December 007 and efinancial qual to one crisis. from January 2008 to July 2013.

This sample size and the fraction r is equal to 0.10. Given our sample size of 2,406 trad prior to the outbreak of the2global Homm and Breitung 2012;0 Bohl, Kaufmann and Stephan 2013), currency (Betten (5) dummy coding enables to examine the aall bnormal performance and factor exposures prior to and It is also consistent withus the fact that of the renewable window roughly covers t2013) he first sample year and (Phillips therefore a sufficient umb commodity (Gutierrez and housing markets andyields Yu 2011; Yiu, Yu nand Ji the oindices utbreak opeaked f the global crisis. is also 1consistent with the fact that all of the energyafter stock atfinancial the end ofIt 2007. to e nsure e stimation e fficiency.

T he w indow s ize e xpands b y o ne o bservation a fter 2 renewable energy stock indices peaked at the end of 2007.

Note that forSthe test the recursive ADF rolling statistic SADF is defined as: the window length is BUBBLE DETECTION TESTS not forward expanding but held constant with4 r equal to Bubble Detection Tests 0.20.

To u ncover p otential e xplosiveness i n t he d eflated p rice t ime s eries o f r enewable e nergy s tock i ndices, SADF ( r ) = sup ADF .

0 s To uncover potential explosiveness in the deflated price time r0 , 1] we use recursive and rolling supremum Augmented Dickey-­‐Fuller (SADF) tests (Phillips, Wu and s∈ Y[u

series of renewable energy stock indices, we use recursive InShi the presence of multiple bubbles, the GSADF test 2011) as well as the generalized SADF (GSADF) version introduced by Phillips, and Yu (2013). These and rolling supremum Augmented Dickey-Fuller (SADF) proposed and Yu (2013) is lassumed toforward be a expanding but Note that by for Phillips, the rolling Shi SADF test the window ength is not right-­‐tailed unit root tests have proved useful in detecting exuberance or bubble-­‐like behaviour in

powerful rb ased equal 0.20. method. The test procedure is designed to financial time series and are applied to daily real price data. The SADF tests more are on to the consistently assumption that asset prices follow a random walk and thus contain a unit root. Exceptions to detect the rule the existence of periodically collapsing 1 We also theupward Quandt-Andrews unknown breakpoint test (Andrews In the presence are suse trong departures from fundamental values, which can lead to explosiveness in the of multiple bubbles, the GSADF test proposed by Phillips, Shi and Y 1993; Andrews and Ploberger 1994) torformally check whether there a underlying price time series.

The ecursive SADF test estimates the is conventional to Abugmented e a more Dpickey-­‐ owerful method. The test procedure is designed to consistently dete 2 SADF-type tests have been employed to test for speculative bubbles structural change in all of the five-factor model’s parameters. The tests Fuller (ADF) regression repeatedly by using a forward expanding sample sequence: periodically c ollapsing bubbles.

Given the and findings in Bohl, Kaufmann and Stephan reveal that a breakpoint occurs in February 2008 for the US stock index, in equity (Phillips, Wu and Yu 2011; Homm Breitung 2012; Bohl, in January 2008 for the European stock index and in July 2008 for the Kaufmann and Stephan 2013), currency (Bettendorf and Chen 2013), comparable sample period, we conjecture that there is mainly one single extended P global stock ease φ of Δ comparability, we date only one common commodity (Gutierrez 2013) and housing markets (Phillips and Yu 2011; Δyt indices. = µ + δyFor (4) there t −1 + p yt − p + ε t , ups. However, could also be several temporary episodes of explosiveness in t breakpoint in January 2008p =in Yiu, Yu and Jin 2013).

1 the regression specifications.

3

∑

Recall that the recursive SADF test fixes the start points of the subsamples on the f

where yt is the daily real alternative energy stock price index, Δ stands for the first difference the total sample, while the rolling approach keeps the window length constant. By

6 • CENTRE FOR INTERNATIONAL GOVERNANCE INNOVATION operator, and µ , δ and φ are regression coefficients. The error term ε p

t

GSADF procedure extends the subsample sequence by changing both the start poin is independent and

points of the subsamples over a feasible range of flexible windows. The GSADF test identically distributed with zero mean and constant variance. To determine the optimal lag length P in each subsample regression, we follow the procedure suggested by Campbell and P erron ( 1991). backward expanding SADF test repeatedly for varying end points Tr with r ∈ [r ,

0

s∈[ r0 , 1]

s

(5)

SADF (r0 ) = sup ADFs .

(5) ALTERNATIVE ENERGY STOCK PRICES? s’ EXPLOSIVENESS [ r0 , 1]our sample size Note of 2,406 trading days, the iDROVE nitial r0 is equal to 0.10. Gs∈iven that for the rWHAT olling SADF tTHE est MID-2000 the window length is not fIN orward expanding but held constant with

EVIDENCE FROM US, EUROPEAN AND GLOBAL INDICES rst sample year and therefore yields a sufficient number of observations r equal to 0.20. that fsor the rsolling test trhe waindow length is ence, nsot forward xpanding ut thhe eld constant with sample ize abnd he to e0ach .10. iven Ho ur ample size of 2e,406 trading db ays, initial cy. TNote he window ize expands y SotADF ne fraction observation fter pGass.

0 is equal bubbles. Given the findings in Bohl, Kaufmann and set to zero. While the empirical supremum test statistics r

e qual t o 0 .20. tcomparable he presence of therefore multiple bubbles, GSADF est proposed with

by Phillips, Shi and Yquantiles u (2013) iof s athe ssumed window roughly covers sample year and yields a sufficient number oto f otbe bservations s defined as: Stephan (2013) forthe In afirst sample period, we the have compared the right-tailed

to ensure estimation efficiency.

The window sowerful ize expands y one observation after each pass. iHs ence, to mainly be a m ore mbethod.

The test procedure to test consistently etect the existence of conjecture that there is one psingle extended phase distribution ofdesigned simulated statistics,dthe date-stamping In the presence f m ultiple ubbles, t he G t est p roposed b y P hillips, S hi and Yu is(2013) is on assumed the ro ecursive SADF sbtatistic is defined aSADF s: (5) of price run-ups. However, there could also be several procedure based expanding sequences of ADF periodically collapsing bubbles. Given the findings in Bohl, Kaufmann and Stephan (2013) for atest

to be a more powerful episodes method. of The test procedure is designed to consistently detect the corresponding existence of critical value sequences temporary explosiveness in the price indices. statistics.

The SADF (r0 ) = sup ADFs .comparable sample period, we conjecture that there i s m ainly o ne s ingle e xtended p hase o f p rice run-­‐ sis ize athe nd is tehe qual to h0eld .10. iven ur w sample size oof f 2,406 trading days, (5) t2013) he initial r0SADF [ rthe ,recursive 1] fraction Recall that test fixes the start are periodically csample ollapsing bs∈fubbles. findings in Boohl, Kpoints aufmann and cv(s) Stephan (obtained for by a first running 5,000 Monte Carlo iven F test the window length not orward eGxpanding but cGonstant ith ups.

Hobservation owever, there could also bufficient e several temporary episodes of explosiveness in the price the subsamples on the first the total sample, simulations the varying subsample windows andindices. then window roughly covers first sample ear atnd tof herefore yields ao sne number of observations comparable sample por eriod, we tShe cADF onjecture tyindow hat here iis s nm ainly single eut xtended pfor hase of price run-­‐ Note t hat f t he r olling t est t he w l ength ot f orward e xpanding b h eld c onstant w ith while theestimation rollingefficiency. approach window length the quantiles the resulting of Recall that the size rthe ecursive tbservation est fixes atconcatenating he esach tart points of tright-tailed he subsamples on the of first observation to ensure The wkeeps indow expands bSy ADF one o fter pass.

Hence, r etqual to By .20. ups. However, here c0ould astatistic lso be ithe several temporary episodes of the explosiveness in the price indices. constant. contrast, GSADF procedure extends simulated the recursive SADF s dSefined ass: ubbles, the GSADF test proposed by Phillips, and Yample, u (2013) s assumed the thi otal wihile the rolling approach keeps distributions. the window length constant. By contrast, the subsample sequence by changing both the points the recursive test fixes tthe sSADF tart peest oints ostart f otf he shillips, ubsamples ou n (2013) the fiirst observation of In the presence oSf ADF m b ubbles, he G t p roposed b y P S hi a nd Y s od. TRecall he test tphat rocedure is designed to ultiple consistently d etect t he xistence procedure extends the range subsample sequence abssumed y changing OF both the start points and the end SADF (end r0 ) = points sup ADF .GSADF CONSTRUCTION IDIOSYNCRATIC and the of the subsamples over a feasible s (5) to be ore K pstaufmann owerful method.

The test (p2013) is to clonsistently detect the Bey xistence of the sample, hile approach krocedure eeps the window ength constant. contrast, es. Gthe iven total the findings in aw B mohl, a nd Stephan for a designed ∈ [he r , 1] rolling points of tthe he subsamples over a feasible range of TIME flexible windows. The GSADF test implements the of flexible collapsing windows.

TheGiven GSADF test implements ohl, Kaufmann athe nd Stephan (2013) for a procedure he bssubbles. ubsample sequence y changing both the PRICE start points and SERIES the end we cGSADF onjecture that periodically there ies xtends mainly otne ingle extended pfindings hase oif n b pBrice run-­‐ backward expanding SADF test repeatedly for varying Note that for sample the rolling SADF test the weindow not forward eepeatedly xpanding but fh onstant wrnd ith points Tr with r ∈ [r , 1] and makes comparable period, we conjecture that ltength here iis ainly one sringle extended peld hase of price e un-­‐ backward xpanding Ss m ADF test or vcarying 2 0 so bpoints e several t emporary e pisodes o f e xplosiveness i n t he p rice i ndices. of the owith ver a rf2aeasible range of flexible windows.

The GBohl, test implements the 2 (2013) and Stephan report substantial end points Trthere ϵ[r and makes inferences rs ubsamples equal to 0.20. ups.

However, could lso 0b,1] e several temporary episodes of ebased xplosiveness iSADF n the pKaufmann rice indices.

2 F test fixes the start p oints o f t he s ubsamples o n t he f irst o bservation o f inferences b ased o n t he s upremum v alue o f t he b ackward S ADF s tatistic s equence d enoted by evidence of explosiveness in the real prices of German on the supremum value of the backward SADF statistic Recall that tShe recursive test fixes ftor he svtart points f the subsamples the frirst of makes backward expanding ADF test SrADF epeatedly arying eond points and ∈observation [r , 1]ssumed Tr own ith In the presence of lm ultiple cb ubbles, the GcSADF test tphe roposed by Phillips, 2Shi and Yu 2(2013) 0 is aenergy alternative stocks.

The goal of the present study olling approach keeps t he w indow ength onstant.

B y ontrast, the total sample, while the BSADF rolling approach keeps window length constant.

By contrast, the Tdhe G SADF test statistic is thus given by )t. .he GSADF test sequence denoted by r2o ∈f [prrocedure ,1] ( rb0ackward 0 he to be abo y mcore powerful method.

Talue he tpest is The to consistently dsetect existence of inferences based the supremum SADF statistic equence deenoted y to tahe examine to which the explosive price e subsample sequence hanging both tthe tart oints and the eesigned GSADF pn rocedure extends he ssvubsample stequence by cnd hanging both the start is points nd the nd the bextent 0

0

statistic is thus given

by Given the findings in Bohl, Kaufmann and Stephan (2013) for a periodically collapsing bubbles.

ample ritical alues for for the the Sby SADF ADF nd GGSADF SADF tests tests re oobtained btained from MMonte ont can bevvalues explained rising crude oilaare prices andfrom Finite ssample ccritical aand of the ubsamples ver as tatistic feasible of flexible wb indows. he GSADF behaviour tFinite est implements the er a fBSADF easible range points of f(lexible wsindows.

The GSADF test range iimplements the GSADF toest s thus given y one r2sTingle r2 ∈[ r0comparable ,1] r0 ) . The sample period, BSADF we conjecture t hat t here i s m ainly e xtended p hase o f p rice r un-­‐ with 55,000 ,000 rreplications. eplications. TThe he ssample ample ize TT i s is 2003 hosen to ccorrespond orrespond to tan the 22,406 ,406 trad tra between and 2007.

Such )or =varying supend pADF , with r2 ∈the with ssize cchosen to to he expanding SADF oints rTr akes bull market [r0 , 1prevailing ] and m est repeatedly for vbackward arying end points wtest ith repeatedly and makes [rr20(,r10t]femporary 2 ups.

However, there cTr ould everal of e1xplosiveness iperiod n the po rice indices. us to 2 s∈ 2 also bre r1∈[ 0, re2pisodes −r0 ] analysis enables determine the explosiveness which hich ddetermines etermines the initial initial wwindow indow length lengt f investigation. investigation.

The he fraction rr0, , whether (6) w t he period o f T f raction 0 inferences ased on the r2sSupremum value f sttart he bpackward ADF statistic sequence denoted by Recall that bthe recursive ADF stequence est fixes the oints subsamples on the first observation of and can thus emum value of (the statistic doenoted by of tShe was sector-specific bewregarded as raolling genuine (6) BSADF )b=ackward sup SADF ADF , the r ecursive a nd g eneralized p rocedures, hereas for for the the regressions rr i s isf r2 r0BSADF r the r ecursive a nd g eneralized p rocedures, whereas rolling regressions 1 (r ) . The GSADF test statistic is thus given by By contrast, the 0−,1 ∈[r0s2 ,∈ample, r[2rb r0] ] 0 while the rolling approach keeps the window length constant. speculative bubble.

Besides the strong sensitivity to stock DF test statistic is tthe hus total gr1iven y suggested bby y PPhillips, hillips, SShi hi aand nd YYu u (2013), (2013), the the ddata ata ggenerating enerating pprocess rocess pecified aas s GSADF (r ) = sup BSADF (r ) . suggested is is sspecified

GSADF procedure extends the subsample 0sequence by changing broth t0he start pmarket oints and fluctuations the end 2 and the positive correlation with oil (6) r2∈[ r0 , 1] with ddrift rift ccomponent omponent TT-­‐1-­‐1 nd the the DF egressions’ lag oorder rder PP i s is sset et to to ero. WWhile hile th r2 a and AADF rregressions’ lag zzero. twith BSADF ( r ) = sup ADF points of rt2 he subsamples over r1a f,easible range of flexible windows. The GSADF est i mplements the studies 0 prices, previous identify the prices of technology r2 r1∈[ 0, r2 −r0 ] ( r )and supremum ttest est sstatistics tatistics hhave ave to to bbe e ccompared ompared wwith ith the the right-­‐tailed right-­‐tailed qquantiles uantiles oof f the th = sup BSADF . that the recursive rolling SADF tests are nested Fr1 , GSADF (rNote supremum 0) r 0 2 backward expanding SADF test repeatedly for varying end points Tr2 with r2 ∈stocks akes [r0 , 1] and asmtaest influential variableprocedure of cleanis energy stock (6) third ∈[ r0 , 1] simulated tatistics, the he ddG ate-­‐stamping ased xpanding seque equ determines in ther2GSADF procedure.

The fraction r0, whichand that simulated test sstatistics, the ate-­‐stamping procedure Tihe s bbased oon n eexpanding s Note t he r ecursive r olling S ADF t ests a re n ested i n t SADF p rocedure. f raction , r 0 inferences on the supremum value of the backward ADF statistic sequence denoted by prices (Henriques and cSadorsky 2008; Kumar, Managi GSADF (rb0 ased ) = sup BSADF (6) Sto r2 ( r0 ) . statistics. he ccorresponding orresponding ritical vvalue alue ssequences equences re btained bby y first first rur the minimum window length, is again equal 0.10. For cv(s(s)) a are statistics.

TThe critical oobtained cv r2∈[ r0 , 1] tatistic is thus tghe Matsuda 2012; Sadorsky 2012a).

There aretest, alsowsigns GSADF test dslag iven binimum y because which etermines window and length, is again etqual to 0.10.

For the GSADF e set the the BSADF (r0 ) . The Fr2 (rNote r2 ∈[ r0 ,1] test, the GSADF setSADF the order to m zero 0 ) . that the Carlo s imulations f or he v arying s ubsample w indows a nd t hen c oncatenating recursive and rwe olling tests are nPested in the GSADF procedure.

T he f raction , r Carlo simulations for arying subsample indows then oncatenating the ri Finite sample critical values for tthe he SvADF and SADF tests are ow btained from a Mnd onte Carlo scimulations 0 Gbubble, a5,000 crude oil which rapidly

lag order P ttests o that zero bested ecause Shi aof nd Yshort-lived u frraction (2013) she ize ddtdo istributions. istortion is collapsed mallest w Phillips, and Yu (2013) with eplications. sample tshat ize T is s chosen correspond to the 2s ,406 trading days in hen our a quantiles of f tthe he rr0Tresulting esulting simulated imulated istributions.

Note that tShi he recursive and rolling Sshow ADF are nsize idistortion n the PGhillips, SADF pis rocedure.

The , how quantiles o s r2 following theTthe price 2008the (see, which determines w indow length, is again equal to 0.10. For the GSADF est, wsurge e he BSADFtrhe (r0 m ) =inimum sup ADF , wet hich tdin etermines initial wfor indow example, length, is set to 0Tokic .10 for period of investigation. fraction r0 , s r1 length smallest a fixed lag is used.

2when fixed length is is uased. which determines wlag indow length, gain equal to 0.10. For the G SADF test, wge sKesicki et the r1∈[t0he r0 ]inimum 2 −m 2010; 2012; 2010). renewable olling SADF tests re zero nested in the G,PrSADF procedure. raction the recursive eneralized whereas for the rolling regressions r is energies fixed at 0.20. As are 0 , Construction of f pIrocedures, Iw diosyncratic rice ime SSeries eries lag order P tao because hillips, Shi and TYhe u (f2013) srhow that size distortion is and smallest hen a Since Construction diosyncratic PPrice TTime lag order P to zero because Phillips, Shi and Yu (2013) show that size distortion perceived is smallest w(6) hen a aoa suggested

by Phillips, Shi nd Yu (2013), the data generating process is specified as a random walk as long-term substitute for conventional A primary advantage of the SADF tests is that they allow -­‐1 Bohl, aufmann nd tShe Stephan tephan (2013) vidence of f eexplosiveness xplosiveness ini um fixed window length, is ais 0BSADF .10. For (trhe GSADF test, we set the with drift T a a and nd ADF regressions’ lag oeport rder P iss s substantial et to zero. While the empirical Bohl, KKcaufmann (2013) rreport eevidence ond GSADF (re0 )qual = is sup lag length used. fixed lgain ag length uto sed.

0) .

A prrimary advantage of the SADF tof ests is energy that tomponent hey allow dcompared ate-­‐stamping tubstantial he oalternative rigination aenergy 2 sources, iteftor isbe likely that for date-stamping origination and termination supremum test statistics have o with Tthe he prices right-­‐tailed qtuantiles of the ds istribution r2∈[ r0 , 1] German alternative lternative nergy stocks. tocks. goal oal oof f tof he present resent tudy is too tf o eexamine xamine the the e is smallest Phillips, Shi and Yu (2013) show that size the distortion when a German a e nergy s T he g he p s tudy i s simulated test statistics, the t dhe ate-­‐stamping rocedure is sbupremum ased oby n expanding equences of Aprices.

DF test have been initially driven soaring termination of tehey xplosive pdrice behaviour.

Provided that full spample tsrude est sooil tatistic explosive price behaviour. that thefor full sample A primary advantage of the SADF tProvided ests is that allow ate-­‐stamping the stocks oexplosive rigination arice nd p behaviour ehaviour can an be e eexplained xplained b y r ising c il p rices a nd t he p reva explosive p rice b c b b y r ising c rude o il p rices a nd t he p revail statistics.

T he c orresponding c ritical v alue s equences

a re o btained b y f irst r unning 5 ,000 M onte cv (s ) A primary asupremum dvantage of test the statistic SADF trolling ests iSs ADF that they llow in for date-­‐stamping the he fraction origination and Note that the and tests are right-sided naested tfhe SADF psrocedure. , the rl0ocate However, after oil bubble burst inus July 2008, the the critical termination of recursive explosive pexceeds rice bexceeds ehaviour.

Prrovided that the ull G sample upremum tsest statistic the ight-­‐sided critical value, it is pTossible o tahe eSpisodes oaf nd ethen xuberance. instance, Carlo imulations ftor varying subsample indows concatenating he ror ight-­‐tailed between 003 and nd 007.

Such uch awan n aanalysis nalysis nables tto etermine hether the the eexp x between 22003 22007. eenables us to dFdetermine wwhether positive correlation have broken down.

In order to Besides termination o f e xplosive p rice b ehaviour.

P rovided t hat t he f ull s ample s upremum t est tatistic quantiles oest, f tihe rw esulting sts imulated dmay istributions. value, it is possible to locate episodes of exuberance.

For which d etermines t he m inimum w indow l ength, i s a gain e qual t o 0 .10.

F or t he G SADF t e s et he exceeds t he r ight-­‐sided c ritical v alue, i t i s p ossible t o l ocate e pisodes o f e xuberance.

F or nstance, sector-­‐specific a nd c an t hus b e r egarded a s a

g enuine s peculative b ubble.

SADF tests is that they allow for date-­‐stamping t he o rigination a nd sector-­‐specific a nd c an t hus b e r egarded a s a

g enuine s peculative b ubble.

B esides t using the recursive Swe ADF ttest we can compare the sequence of ssensitivity ubsample Apatterns, DF coefficients ADF such dynamic we allow for s instance, using can compare lag order to zc ero bthe ecause Phillips, nd Yu test (he 2013) how hat size distortion iConstruction s smallest w hen aor m exceeds Ptrovided he right-­‐sided ritical vtalue, iSs hi paossible tso lsocate oADF f ecapture Fdiosyncratic instance, using he Prtecursive ADF recursive est we ict an cSADF ompare equence of e spisodes ubsample cxuberance. oefficients osf stock Itock rice Time Sa eries ADF sensitivity arket fluctuations fPluctuations and nd the the ppositive ositive ccorrelation orrelation wwith ith ooil il pprices, rices,p e behaviour. tthat he full sSample supremum test sttatistic s market sensitivity tto o time variation in(2013) our linear multifactor model.

Bohl, vKalue aufmann nd Stephan report evidence of explosiveness in the rWhether eal prices of the sequence coefficients ADF with thecritical fixed lag length of is usubsample sed. with tADF : stocks he corresponding right-­‐tailed saequence cv (s )substantial identify he rices osof f ttechnology echnology tocks s aa t hird tis hird influential ariable of f cclean lean eenerg nerg : subsample with to the cSorresponding critical vFalue sequence cvs(so)f identify tcthe pprices aas nfluential vvariable al value, it tis possible locate episodes of cean xuberance. or instance, using he recursive ADF tright-tailed est wright-­‐tailed e ccritical ompare the ssequence equence ADF oefficients German alternative energy tocks.

TADF he goal of tshe present study to eixamine the extent to can which othe the explosive price behaviour was sector-specific be s corresponding value cv(s): (Henriques a nd S adorsky 2 008; K umar, M anagi a nd M atsuda 2 012; S adorsky 2 012a explosive p rice b ehaviour c an b e e xplained b y r ising c rude o il p rices a nd t he p revailing b ull m arket (Henriques nd Sadorsky 2008; Kumar, Managi and Matsuda 2012; Sadorsky 2012a primary advantage of the SADF tests is that they aADF llow for the investigated o rigination aand t we with can compare tA he sequence of subsample A DF coefficients date-­‐stamping bySuch orthogonalizing the price serieswwith between 2003 and 2007. an analysis enables us to determine whether time the explosiveness as the corresponding ritical value s:equence cv }(s rˆe = inf s ≥ r0 {osf :eright-­‐tailed ADF cvrice (ˆs)cb}=ehaviour. , inf rˆf = inf ADF <s cv ((ss)sample . ), : srˆupremum (7) signs a s s:hort-­‐lived hort-­‐lived rude ubble, which hich ccollapsed ollapsed rapidly following following the price s s >p srovided ≥ rˆe:{sADF st> signs oof f atatistic ccrrude rapidly termination xplosive P t hat he f ull t est s r s cv ) = inf s ADF cv (oosil ail s )bab ubble, . swpeculative

sector-­‐specific a nd can thus egarded genuine bubble. Besides the strong (7) respect aforementioned systematic risk factors andthe price su e s ≥ r0 s f s ≥ rˆe to the sbe < -­‐tailed critical value sequence cv (s ) :

for e xample, T okic 2 010; 2 012; K esicki 2 010).

S ince r enewable e nergies a re perceiv sensitivity to isnstance, tock T mokic arket f2 luctuations and the positive 2correlation with oril enewable prices, previous studies are for example, 010; 2012; Kesicki 010). Since energies exceeds the right-­‐sided critical value, it is possible to locate episodes (7) of exuberance.

For analyzing the remaining idiosyncratic component.

To perceive identify t]he pnd rices of tcechnology stocks ae s anergy third influential variable olf ikely clean etnergy stock prices ˆ The origination date of explosive price behaviour is eventually derived from τˆe then

a t he = [ T r substitute f or onventional s ources, i t i s hat p rices o f a lternative ne e onventional sources, it is likely tseries, hat prices of afirst lternative eener the recursive compare sequence f subsample csubstitute oefficients for rˆ = inf using s : ADF > cvS(ADF s) t,est rˆwf e =can inf : ADF ( s ) . ADF construct ADF (7) Menergy s2 008; sector-specific as cedaily price we s ≥ rˆe sthe s < ocv (Henriques and Sadorsky Kumar, anagi and Matsuda 2012; Stime adorsky 2012a). There are also ˆ ˆ

The o rigination d ate o f e xplosive p rice b ehaviour i ventually d erived f rom

a nd t he s) , erˆf = infss≥≥ rrˆe0 expression s : ADFs τ<sˆ cv ( s ) . = [ T r ] (7) been i nitially d riven b y s oaring o il p rices.

H owever, a fter t he o il b ubble b urst i n J ul ˆ e the etshe the subsequent collapse date. cv (s ) : been driven y ubble, soaring il prices.

However, after il 2b008 ubble signs of ianitially short-­‐lived crude oil which ocollapsed rapidly following price urge oin (see, burst in July f = [Trf ] y ields with torigination he corresponding right-­‐tailed critical value sequence estimate the rawbbindex returns’ time-varying exposures to The date of explosive price behaviour is for e xample, T okic 2 010; 2 012; K esicki 2 010).

S ince r enewable e nergies a re p erceived a s a

l ong-­‐term correlation m ay h ave b roken d own.

I n o rder t o c apture s uch d ynamic s ensitivity may have broken down. In order to capture such dynamic sensitivity ppaa ˆ f = [Tis rˆef ]ventually acorrelation setcˆof systematic factors: expression yields the subsequent ollapse d] ate. substitute for conventional end nergy she ources, it is likely that prices of alternative energy stocks have ˆ The o rigination d ate o f e xplosive p rice b ehaviour d erived f rom a t = [ T r time v ariation i n o ur l inear m ultifactor mm odel.

Whether hether the xplosive rice bbehavio ehavi sive price behaviour ventually derived f(rom the the e driven Trˆse≥]rˆ a{snd time variation our linear mHultifactor odel.

W he eexplosive pprice rˆe i=s einf : ADF s)}, τˆrˆef == [inf : ADF ( s )} .

(7) bien been initially y soaring oil prices. owever, after the oil bubble burst in Jtuly 2008 the positive eventually from and expression s ≥ r0 {sderived s > cv s < cv e

specific c an b e i nvestigated b y o rthogonalizing t he p rice t ime s eries w ith r espect to correlation c man ay hb ave roken down. In ob rder capture such dynamic sensitivity we allow for respect to specific e ibnvestigated y oto rthogonalizing the price ptatterns, ime series with ˆ f = [Tdrˆate. yields tthe he subsequent subsequent collapse collapse date. date. he sexpression ubsequent collapse time variation in our linear multifactor model. Whether the explosive price behaviour was sector-­‐ f ] yields 5 (8) ystematic rrisk isk factors factors aand nd then then aanalyzing nalyzing the the remaining remaining idiosyncra idiosyncr The origination date of explosive price behaviour is eventually derived from τˆe aforementioned abnd the ssystematic =aforementioned [Trˆec]an specific e investigated by orthogonalizing the price time series with respect to the construct aa ds ystematic daily aily ssector-­‐specific ector-­‐specific price rice time series, we e first first eestimate stimate the raw index index rer aforementioned risk factors and then analyzing the reries, emaining idiosyncratic component.

To construct p t ime s w t he r aw Finite sample critical values for the SADF and GSADF tests expression τˆ f = [Trˆf ] yields the subsequent collapse date. construct a daily tso pystematic rice time series, we first estimate the raw index returns’ time-­‐varying exposures et oof f ssystematic factors: exposures to ector-­‐specific aa s set factors: (9) are obtained from Monte Carlo simulations with 5,000

{

}

{

{

}

}

}

{

{

}

}

τ

τ

τ

τ

replications. The sample size T is chosen to correspond 5 2,406 trading days in our period of investigation. to the 5 The fraction r0, which determines the5 initial window length, is set to 0.10 for the recursive and generalized procedures, whereas for the rolling regressions r is fixed at 0.20. As suggested by Phillips, Shi and Yu (2013), the data generating process is specified as a random walk with drift component T1 and the ADF regressions’ lag order P is

5

exposures to a set of systematic factors:

RR α=t α +αβt1+ β5t Oilt+ +β ββ6t Oil R +ββ1Market +(8) t = t Market t + β 2t Market t −β 1+ 3tTecht + β t + t −1 + εtt −,1 + ++ββt −13Tech ββ5Oil t= t Market t ++β 2Market tβMarket t −41tTech t+Tech 4Tech tTech t Oilt++ββ 6Oi tO t

t

1t

t

α t = α t −1 + ν t , ν t ~ N (0,σν2 ) ,

2t

t −1

3t

t

4t

t −1

(10) 5t

t

6t

(9)

N((00,σ ,σν2ν2)), , the returns of a broad stock ααt t ==ααt −t −1and ννt t ~~N 1++ννt ,t ,Tech Market represent

β = β + η , η ~ N (0,σ ) .

(10) market index andη a technology stock index, respectively, 22 ,aσ ηbηroad Market Tech r+ epresent r~ eturns f σ stock market index and a technology stock index, ββktkt ==and ββOil ηηktkt, , ηηtkthe NN((0o0,spot )). .price kt ~ while the variations of crude oil.

11+reflects ktkt−− kk of crude respectively, while Oil reflects the spot price variations oil. Although the alternative energy Although the alternative energy stock indices comprise stock indices comprise mostly liquid and frequently traded stocks that quickly assimilate market-­‐wide Market aand nd ech rrepresent epresent tnhe eturns broad road tock mm arket index aand nd aa t echno techno Market TTech rreturns oof f ian a tbhe sstock arket index news, delayed information processing otr he on-­‐synchronicity trading of stocks and explanatory mostly liquid and frequently traded stocks that quickly kt

kt −1

kt

kt

2

k

factors could arise w dw ue to time zrone differences. e pot therefore av pply Dimson’s (o 1979) correction respectively, hile reflects eflects the W rice variations ariations of f ccrude rude il. aAnd Although lthough the the aal respectively, hile OOil il the sspot pprice ooil. additionally include lagged values of the right-­‐hand side exogenous variables. stock i ndices c omprise m ostly l iquid a nd f requently t raded s tocks t hat q uickly ssim stock indices comprise mostly liquid and frequently traded stocks that quickly aassim news, ddelayed elayed information information processing rocessing oor r nnon-­‐synchronicity on-­‐synchronicity in in the the trading trading oof f sstocks tocks news, Martin T. Bohl, pPhilipp Kaufmann and Pierre L. Siklos •7 factors ccould ould aarise rise ddue ue tto o time time zzone one ddifferences. ifferences. W We e therefore therefore aapply pply DDimson’s imson’s (197 (197 factors 6 additionally include include lagged lagged vvalues alues oof f the the right-­‐hand right-­‐hand sside ide eexogenous xogenous vvariables. ariables. additionally

CIGI Papers no. 36 — July 2014

assimilate market-wide news, delayed information DATA processing or non-synchronicity in the trading of stocks We examine five widely followed alternative energy stock and explanatory factors could arise due to time zone indices differences.

We therefore apply Dimson’s correction The time-­‐varying coefficients of the state-­‐space model (1979) are assumed to follow a pure random for walk the and sample period from January 2004 to July 2013.

For ease and additionally include lagged values of the right-hand are denoted by α t and β kt with k = 1, …, 6. The normally distributed error terms ε t , ν t and η kt are of comparability, the sample begins when data for all indices become available. The time series are side exogenous variables.

2 2 independent of each other with zero mean and variances denoted by σ ε , σ ν and ση2 . Equation (8) denominated in US dollars. Unless otherwise stated, we represents the measurement equation, of whereas the transition model equations (9) and (collect 10) describe the The time-varying coefficients the state-space are data from Thomson Reuters Datastream. evolution of to the follow unobserved state random variables. W e use and the Kalman filter technique to estimate the time-­‐ assumed a pure walk are denoted e time-­‐varying cаoefficients f the sktate-­‐space m6.odel assumed to distributed follow a pcure random walk and varying coefficients.

The K=alman filter is aa rre ecursive procedure for omputing the oFor ptimal estimates of the US market, we focus on the WilderHill Clean by and βkt owith 1, …, The normally t and

w ith k

=

1 , … , 6 .

T he n ormally d istributed e rror t erms ,

a nd

a re e denoted berror y αst tate ε ν the vβ ariables f or e ach p eriod t , c onditional o n t he i nformation s et a vailable u p t o t ime t

( Kim a nd η Energy Index, which tracks the performance of clean terms ԑ , ν and η are independent of each other t t kt kt t t kt 2 2 2 the 2 parameters 2 energy companies listed on major US stock exchanges.

Nelson 1 999; D urbin a nd K oopman 2 001).

W e f irst e stimate o f t he m odel ,

a nd σ σ and variances denoted by nd ση . Equation (8) dependent owith f each zero other mean with zero mean and variances denoted by σ ε , σ ν aand ε ν In July 2013, the index consisted of 51 stocks. We obtain 2 via maximum likelihood and then dtransition erive the efiltered values f t(he state variables σmη easurement presents the equation, whereas the quations (9) aond 10) describe the α t and β kt . We . Equation (8) represents the measurement equation, The time-­‐varying coefficients of the state-­‐space model are monthly assumed to follow a pure rdata andom walk and from the New York Stock total return directly olution of the unobserved state variables. equations We puredicted se the Kvalman technique to estimate the tthe ime-­‐ specify the ithe nitial one-­‐step-­‐ahead alues foilter f (10) the state variables by taking OLS estimates whereas transition (NYSE) The European Renewable nd and ith k =describe 1, …, 6. The normally Exchange distributed error terms ε t Euronext. , ν t and η kt are are denoted by α t a(9) β or kt w rying coefficients.

Kalman filter a rtecursive rocedure computing We the ouse ptimal estimates of from evolution a Tshe tatic regression ois ver he first 2p52 trading ays. the of the unobserved state fdvariables.

Energy Index (ERIX), 2 2 2which incorporates the 10 largest and ach other with zero mean aund variances denoted e state variables for each period t, cindependent onditional on otf he information available p to time t (Kim and by σ ε , σ ν and ση . Equation (8) the Kalman filter to eestimate thesaet time-varying most liquid from the wind, solar, biomass and time-­‐varying oefficients of ttechnique state-­‐space odel re ssumed tare o ffactor-­‐related ollow a2 pture random wcompanies alk and alk The The aclternative energy sche tock index return cm an be da ecomposed into contribution 2rfeturn t ime-­‐varying oefficients o f t he s tate-­‐space m odel a a ssumed o ollow a pure random and the elson 1999; coefficients.

Durbin and Koopman 2001).

We ffilter irst estimate the parameters ow f thereas he model represents the miseasurement equation, the σtransition equations (9) and (10) w describe The Kalman a recursive procedure ε , σ ν and water energy sector, is employed to cover the European and s ector-­‐specific component. ach ntrading day, the idiosyncratic part is computed the η bwy ith koptimal 1, β …, Fo6 or . tith Tehe dhe istributed eW rror terms aas nd are nd denoted bfor y are αacomputing β ktand d a=erive nd w =vormally 1alues , …of , o6the . stTtate ormally d istributed eε. rror erms re the time-­‐ dnd enoted α tevolution εsum 2 f he uknobserved vnariables. e α use tnd he βKalman technique estimate t a t , νfttilter η kt atotal the estimates state variables tkt, ν t ato kt ia maximum l ikelihood t hen t he f iltered f he s tate v ariables

a W e market.

Both price and return data are provided by t kt ηe t v and the residual (see Elton eat pal.

1993) and then of the estimated aolpha coefficient αˆ t m εˆtf ilter ime-­‐varying coefficients f the state-­‐space odel re ssumed to fiollow ure andom wfor alk caomputing nd varying coefficients.

Tahe Kaalman s aset recursive p2rrocedure optimal estimates of 2 22 2 2 t(he for each period t,eredicted conditional on the information ,

a nd .

E quation 8) ependent o f e ach o ther w ith z ero m ean a nd v ariances d enoted b y σ σ σ the index proprietor Société Générale.

Our international ecify t he i nitial o ne-­‐step-­‐ahead p v alues o f t he s tate v ariables b y t aking t he O LS e stimates independent o f ach o ther w ith z ero m ean a nd v ariances d enoted b y ,

a nd .

E quation ( 8) σ σ σηk up to time t (Kim and ε oεn tν, he over ime sandom t, hat: nd k =rthe uch 1, …

6w . vTalk he annd ormally distributed terms and εηηktk aν re e denoted bcompounded y α t ato βup tate ariables for each period t, ecrror onditional set available t ν tinformation odel are assumed follow ato ith pture kt w tst(Kim and stock indices comprise the S&P Global Clean Energy om a static ravailable egression over the ftime irst 252 rading d ays. Nelson 1999; Durbin and 2transition resents he easurement w hereas tthe tdw ransition erσ quations (9) a2nd (10) the easurement quation, hereas t2he quations (d9) nd (10) the σ 2 , σ 2 and ormally dtistributed terms , Nelson aean nd are νodel 1η urbin afnd Kparameters oopman 001).

W2w e fnd irst ee stimate the pescribe arameters odf escribe the model Koopman 2001). estimate the Teorror ng coefficients oach f the state-­‐space m are assumed o the ollow a pure andom aσ nd , of aalk . Equation (a8) dependent of m erepresents ther wthe ith em zεquation, ero m a999; nd veDariances enoted by σ tWe t first kt ε ν Energy Global Innovation ε ν Index, the WilderHill New η k * ˆhe evolution f t. he tate ariables.

W e ε u, se the Ktηalman technique tto estimate time-­‐ Petw =nobserved + ×nnobserved Pv02*ariables.

(11) lution odf the uith sTεtate We vulikelihood se the Kaaalman filter echnique tvo estimate he time-­‐ the 2 +α e nergy c.van b e istributed ds(ecomposed into factor-­‐related rthe eturn cfilter ontribution 2ormally t6 tr)u =s(tock 1 1, , σ …oνˆi, 2ndex dmaximum error terms a(nd re y αalternative ia m aximum then iltered alues the of and the state nd β kt . WAlternative e α t aGlobal riances enoted b∏ y kσ ae nd 8) σeturn Index thevariables Ardour Energy Index t and t νand td erive kt m kt a presents tβhe easurement quation, w hereas the transition end quations 9) then and (f10) describe model and via likelihood η E quation ησ. t =1 ε

d a sector-­‐specific component.

For feilter ach trading day, filter the idiosyncratic pfart i s c omputed a s t he s um varying coefficients.

The Kaalman i2s rocedure a r2ecursive p rocedure f or c omputing t he o ptimal e stimates of Clean Energy Index contains ying c oefficients.

T he K alman i s

r ecursive p or c omputing t he o ptimal e stimates o f 2 Composite.

The S&P Global olution f tderive he unobserved state ariables. e the use ostate he filter tαechnique eof stimate the time-­‐ by taking the OLS estimates them(9) filtered values β to .(8) specify the initial pσredicted values the state variables he transition equations and (10) describe tW he , variables . Eand quation f each oother w ith zero ean and vvariances dof enoted btne-­‐step-­‐ahead y σKε alman σ ν aend η he residual (see Elton t ao l. n 1993) anformation nd atkt hen the estimated alpha coefficient αsˆet εˆott , n tate variables or ach period conditional tthe su et vailable to taime t (Kim atraded nd t aftnd state valman ariables fsor eKach ti, s 1one-step-ahead cte00. onditional the information sirading et vailable p atof o tathe ime tu (p Kim nd 30 publicly clean energy companies, The fthe bspecify ase value iinitial s o The oegression rthogonalized pftor rice time series can bptimal e interpreted s the P0p* eriod We the predicted values of from s tatic r o ver he f irst 2 52 t d ays. se t he K ilter t echnique t o e stimate t he t ime-­‐ rying c oefficients.

T he alman f ilter a

r ecursive p rocedure c omputing t he o e stimates of largest measurement equation, whereas the transition esquations (9) maodel nd (10) daescribe the The t ime-­‐varying c oefficients o f t he tate-­‐space a re ssumed t o f ollow a

p ure r andom w alk a nd 2 2 mpounded o ver t ime s uch t hat: 2 2 whereas the WilderHill New Energy Global Innovation Nelson 1 999; D urbin a nd K oopman 2 001).

W e f irst e stimate t he p arameters o f t he m odel ,

a nd σ σ e rocedure computing he e001). stimates the state variables taking the ordinary least squares accumulated value oof ptimal an icnvestment in rf enewable energy sestocks that is ftully rice e variables for vaeariables. ach eriod tu, by onditional oon t… he information et aarameters vailable uep time ta εgainst (odel Kim apσ nd son 1999; Dfor urbin nd Kbptoopman 2βthe W irst the otho f edged he m and ε ν e sputate nobserved state se alman to stimate the and wKith k =e 1effilter , nergy , 6teechnique . stimate he niormally dpistributed rror terms nd are denoted y αWt e ε , ησ t , ν ta a The aktlternative sTtock ndex return can be ime-­‐ decomposed into factor-­‐related kt ν are return contribution * Index is much broader with 96 constituents from 25 2 2 n the Tinformation s et a vailable u p t o t ime t

( Kim a nd from static over 252 2 filter variations in tim he isk afactors.

Tregression he ector-­‐specific terive ime eries weill bvf e alues eflated tested for lson 1T999; Durbin Ks oopman 2l001).

Wse first tthe he ptfirst arameters tdhe m , bσeing nd ients. he Kestimates alman a recursive procedure or ectstimate omputing tshe oFptimal stimates f f ttbhe σisdiosyncratic avnd ia ikelihood afnd hen dnd he fP iltered oodel he ariables e and β kt . aW 2efore σ εtate ν2 v a * via maximum * aximum and a swd component. or et enoted ach to rading do2ay, pThe art . α is s te he shas um a more comprehensive tcomputed η l a nd t hen erive t he f iltered v alues o f t he s tate v ariables a nd W α 2ector-­‐specific 2ero β independent o f e ach o ther ith z m ean a v ariances d b y ,

a nd .

E quation ( 8) σ σ kikelihood σ ˆ ˆ ( ) countries. latter index Pt bles = 1 + α + ε × P .

(11) t ε ν kt days. η ∏ 0 of the mo t t, estimate ttrading he arameters odel , dσ and σe for each conditional the information set available up cˆtombination ime t (Kim 2 an ptp eeriod xplosive root. aN ote tn hat o tνnhe ot nterpret the linear of atnd he αeεˆxogenous vW ariables as and then nd the residual stock (see ee t al.

1993) the eε stimated aplpha cvoefficient α to o satf tate 1 aximum t =m via likelihood nd then dw erive fiiltered alues otf ransition the vsariables nd(10) . aking initial specify the oof ne-­‐step-­‐ahead redicted values the tate variables bβEy lton the Oas LS eitstimates universe, not only includes pure plays but also tta a kttd η represents t he m easurement e quation, w hereas t he e ( 9) nd escribe t he k 2quations 2y taking cify t he i nitial o ne-­‐step-­‐ahead p redicted v alues o f t he s tate v ariables b t he O LS e stimates value, ut β as ktta. he sW et of explanatory factors that t he a forementioned l iterature iltered avnd alues otf rue the fundamental s2tate variables abnd α Durbin Kthe oopman 001).

We first etstimate pe arameters of the model ,

a nd σ σ ε ν compounded o ver t ime s uch t hat: from a

s tatic r egression o ver t he f irst 2 52 t rading d ays. considers companies with a focus on energy efficiency or ecify the initial predicted values of the sW tate variables by taking he 4OLS estimates evolution of the unobserved state variables. e use the Kalman filter ttechnique to estimate the time-­‐ *one-­‐step-­‐ahead ab s koey determinants of rtenewable nergy stocks’ behaviour. m tshe tatic ridentified egression irst 52 rading vsalue is alternative set to 1tver 00.

Tthe he ofenergy rthogonalized price dteays. ime series can rbeturn e interpreted as t he The stock index return can be 0and s e obaf ase tate vPariables y aking the he O LS e2stimates mum l ikelihood t hen d erive t f iltered v alues o f t he s tate v ariables

a nd .

W e α β on the reduction of carbon dioxide emissions. Our third t varying coefficients.

The Kalman filter is a recursive procedure for computing the optimal estimates of kt om a static regression over the first 252 tTrading days. into contribution cumulated vdecomposed alue of aan investment ia n rfactor-related enewable energy sreturn tocks *cthat is e fully hedged aand gainst parice days. * stock index The lternative e nergy r eturn an b d ecomposed i nto

f actor-­‐related r eturn c ontribution international index, the Ardour Global Alternative Energy ˆ ˆ the s tate v ariables f or e ach p eriod t , c onditional o n t he i nformation s et a vailable u p t o t ime t

( Kim a nd (c1an one-­‐step-­‐ahead predicted alues Port f the state y 0taking the OLS e fstimates = + vαariables +dεecomposed .

(11) ∏ t ) ×b*P ial alternative energy stock vindex eturn btFor e into aday, actor-­‐related return contribution ahe esector-specific component. each trading the riations i n t r isk f actors.

T he s ector-­‐specific ime s eries

w ill b e d eflated b efore b eing t ested f or P e a lternative nergy s tock i ndex r eturn c e d ecomposed i nto a

f actor-­‐related r eturn c ontribution 1 tb t =an and a

s ector-­‐specific c omponent.

F or e ach t rading d ay, t he i diosyncratic p art i s c omputed a s t he s um 2 2

Index Composite, consists of 115 clean energy stocks as of t egression over the irst 252 Dtrading ays.

Data Nelson 1omponent.

999; urbin adFnd Kce oopman 2001). dWay, e first the parameters f the model as , σ νsum and σ tε he ecomposed into a ffactor-­‐related return ontribution d aa s sector-­‐specific or ach trading the eistimate diosyncratic art is coomputed idiosyncratic part is the sum of the ector-­‐specific ccomponent.

Faot or ecomputed ach trading das ay, tc he idiosyncratic pestimated art i s cpomputed aeriod s atl. he sum January edxplosive r oot.

N ote t hat w e d o n i nterpret t he l inear ombination o f t he e xogenous v ariables a s We e xamine f ive w idely f ollowed a lternative e nergy s tock i ndices f or t he s ample p f rom July 2013.

Performance data for this index are retrieved ˆ ˆ a nd t he r esidual ( see E lton e t 1 993) a nd t hen of t he e stimated lpha c oefficient α ε * t 2 ay, the idiosyncratic part is computed s tahe sum t vvia m likelihood nd then iltered vlton ot f atl. he s993) tate vaariables We αecome The bf a ase value erive set ttthe o 1f00.

The oalues rthogonalized price time series cnd an β be as the Pesidual σindex e etrue nergy stock ralue, eturn can ecomposed into afactors ifs actor-­‐related reforementioned eturn chen ontribution ˆ ˆ t a 0d a nd t he r

( see E e 1 nd t hen he eestimated lpha c2aximum oefficient α ε kt i. nterpreted ηatlpha e fundamental but aor s bae e α sd et o e xplanatory hat t he a l iterature ˆ ˆ 2004 o J uly 013.

F ase o f c omparability, t he s ample b egins w d ata f or a ll i ndices b

a nd t he r esidual

( see E lton t a l.

1 993) a nd t hen the stimated a c oefficient ε from the online database of Ardour Global Indexes. t t and the residual (see Elton et al. alpha coefficient t dual εˆ (see Elton et al. 1993) aond then compounded ver tt dime uch tvhat: pecific tcomponent.

For each trading ay, tsenominated he idiosyncratic art ibs o cf omputed atherwise the sum accumulated alue oin f time arUeturn n idnvestment r4enewable esnergy sw tocks that fstimates ully hedged against price specify tThe itnitial oeries ne-­‐step-­‐ahead predicted vpS alues tU he sin tate vs ariables by taking tche OLS dis eata available. he ime s a re d ollars. nless o tated, e ollect entified a s k ey d eterminants o f r enewable e nergy s tocks’ ehaviour.

1993) and then compounded over such that: mpounded ver ttime ime such such that: mpounded over that: the residual (see Elton eft al. 1993) and then ed alpha coefficient εˆt t he For the from a αsˆtatic over 2risk 52 tactors. rading dThe ays. from Thomson Rregression euters D atastream. variations in first the sector-­‐specific time series ill be multifactor deflated before performance being tested for evaluation, we follow Pt * w t Tand * * T the literature and use ˆ ˆ Time such that: ( ) P = 1 + α + ε × P . over t (11) an e xplosive r oot.

N ote t hat w e d o n ot i nterpret t he l inear c ombination o f t he e xogenous vmonthly ariables as total returns including 0 t t t * * * energy lternative stock itndex eturn can be dEecomposed a factor-­‐related return contribution m arket, we focus on he WrilderHill Clean nergy Index, into which tracks the p(11) erformance of ata ))×∏ α ++aεˆεU (11) (11) dividends.

Performance measurement is usually based on =P 10P. . t× ˆttthe ˆt tS == ∏((11++For αˆThe (11)

the t rue f undamental v alue, b ut a s a

s et o f e xplanatory f actors t hat t he a forementioned l iterature 0 t =1 and sector-­‐specific clomponent.

For each trading day, part is computed s um eanergy companies isted oen nergy major U S stock exchanges.

In idiosyncratic July p2eriod 013, tfhe index consisted aodata f t5he 1 sto e examine fclean ive widely followed stock indices for the the sample rom January alternative 4 obtain more robust regression estimates. t =1 monthly *

identified a s k ey d eterminants o f r enewable e nergy s tocks’ r eturn b ehaviour.

ˆ t t+o εˆJtuly )×2Pstocks. α . tFhe 0 of W obtain maonthly tsotal ro eturn from tεˆhe N Ylton xchange NYSE) 04 013. or ase ovf comparability, tthe ample bd egins when data or aEll (11) iork ndices ecome ahe nd toirectly he residual eSt tock asbl. eries 1E993) and hen stimated lpha αˆ td Tata t (fsee The bee ase alue s oefficient et rthogonalized pew rice tprice ime can b(te interpreted as the we employ the US, European and P0* icis As our benchmarks, set to1s00.

100.

The orthogonalized The value *base ed rice vTtalue ime series c an b e i nterpreted a s t he e bpase

i s s et t o 1 00.

T he o rthogonalized p rice t ime s eries c an b e interpreted a0 s ltargest he and most P Euronext.

T he E uropean R enewable E nergy I ndex ( ERIX), w hich i ncorporates the d1ata ailable. he time s eries a re d enominated i n U S d ollars.

U nless o therwise s tated, we collect * 0 compounded o ver t ime s uch t hat: time can beThe as ithe accumulated value base value is set to 1v00. rthogonalized rice time seeries csan be itnterpreted aedged s the against price portfolios from Kenneth French’s P accumulated alue ointerpreted f ao n nvestment n prenewable nergy tocks hat global is fully hfactor-mimicking 0 series energy stocks that if s aDfn ully hedged gainst pirice liquid from tahe stolar, nd ater energy is eamployed to cover the 4 om R1euters cumulated value nvestment in w enewable eiomass nergy tocks that is fully hhe edged gainst price Data e is set tof o 00. coompanies Tinvestment he atastream. oirthogonalized prind, rice ime sberies can asb e stocks iw nterpreted as sector, tis P0T*homson The traditional US factors are described in data library. an in renewable energy that fully * T * alue umulated v o f a n i nvestment i n r enewable e nergy s tocks t hat i s f ully h edged a gainst p rice European m arket.

B oth p rice a nd t otal r eturn d ata a re p rovided b y t he i ndex p roprietor S ociété * variations i n t he r isk f actors.

T he s ector-­‐specific t ime s eries

w ill b e d eflated b efore being tested fJor P me series n Pthedged wrill deflated before being for wsidely * be * tested We examine ftive fthe ollowed energy stock indices fand or ftor he sample period from anuary t riations isk factors.

Tˆhe sprice ector-­‐specific ime eries w ill ablternative e racks deflated berformance efore being tested the P Fama French (1993) and comprise the market excess against variations in risk factors.

The ˆ alue n arket, investment i n r enewable e nergy s tocks t hat i s f ully h edged a gainst p rice ( ) P = 1 + α + ε × P .

(11) t r the oUf S aim w e f ocus o n t he W ilderHill C lean E nergy I ndex, w hich t t he p o f ∏ 0tock indices comprise the S&P t t t * Global Clean Energy Index, the WilderHill Générale.

O ur i nternational s 2004 tariables o J uly w 2*013.

Fime or ease of comparability, t he s ample b egins w hen d ata f or a ll i ndices b ecome ations i n t he r isk f T he s ector-­‐specific t s eries

w ill b e d eflated b efore b eing t ested f or P =o 1 f the erxogenous tactors. he l inear c ombination v a s an e xplosive oot.

N ote t hat e d o n ot i nterpret t he l inear c ombination o f t he e xogenous v ariables a s return, the size portfolio SMB, the value portfolio HML t ean energy crsector-specific ompanies ltisted on m ajor U S s tock e xchanges.

I n J uly 2 013, t he i ndex c onsisted o f 5 1 explosive oot.

Note hat w e time do ot i nterpret t he l inear c ombination o f t he e xogenous v ariables a s he risk factors.

The sector-­‐specific tn ime s eries

w ill b e d eflated b efore b eing t ested f or P series The t will be deflated before being time saeries af re edxplanatory enominated factors in US dollars.

U nless otherwise stated, we cportfolio ollect data WML. The construction of t hat the m tonthly he t rue a forementioned ft undamental otal r eturn available. d ata v alue, d irectly but atory W factors l iterature a s

s et o t hat t he a forementioned l iterature and the momentum ocks. e o btain f rom t he N ew Y ork S tock E xchange ( NYSE) e t rue f undamental v alue, b ut a s a

s et o f e xplanatory f actors t hat t he a forementioned l iterature explosive r oot.

N ote t hat w e d o n ot i nterpret t he l inear c ombination o f t he e xogenous v ariables a s * tested for an explosive root.

Note that we do not interpret 4 oot. Note that we d o not interpret tfrom ltinear crolling ombination oDf atastream. the exogenous variables as bofe 252 4check, As abehaviour. robustness we also run regressions with fixed window length trading 4days.

Results The b ase value she et o Thomson 1 00.

The rthogonalized parice time series can interpreted as the Roeuters 0 is gy stocks’ return Pey 4 largest the and global factors is based on 16 and 23 ronext. he Eey uropean enewable Ifndex hich he 10 m ost European eterminants f rsenewable energy sttocks’ eturn behaviour.

entified s kthe dut eterminants f rdEenewable ef nergy tocks’ rincorporates eturn behaviour. arforementioned linear the exogenous variables as and true fundamental ut anergy s toa the set oof e(ERIX), xplanatory factors that the literature mental vTaalue, bidentified as similar avR salue, et acombination os f ekbo xplanatory actors tohat tw he aforementioned literature are very compared Kalman filter approach. accumulated value of abn iomass investment iater n renewable energy setocks that tio s fcully htedged against pcountries, rice developed respectively, and their total returns uid companies f rom t he w ind, s olar, a nd w e nergy s ector, i s mployed over he 4 4 For t he U S m arket, w e f ocus o n t he W ilderHill C lean E nergy I ndex, w hich t racks the performance of fundamental value, but asstocks’ a set of explanatory ey determinants oeterminants f renewable energy stocks’ return behaviour. return ntified as kthe ey dtrue of factors. renewable energy b*ehaviour. are measured in (see Fama and French 2012 for variations in tahe rtisk Tdhe sector-­‐specific ty ime series w ill be edSxchanges. eflated before b2eing tested fUS or cdollars P ropean market.

B oth p rice nd otal r eturn ata a re p rovided b t he i ndex p roprietor ociété t clean e nergy c ompanies l isted o n m ajor U S s tock I n J uly 013, t he i ndex onsisted o f 5 1 factors that the aforementioned literature identified as 7ttotal nérale. Our international srtock indices comprise tnhe Sim &P Global Clean Energy Index, the W ilderHill a more detailed description).

To calculate excess returns, ata Data an e xplosive oot.

N ote t hat w e d o ot nterpret he l inear c ombination o f t he e xogenous v ariables a s stocks.

W e o btain onthly r eturn d ata d irectly f rom t he N ew Y ork S tock E xchange ( NYSE) keyfor determinants renewable energy stocks’ return gy stock indices the sample period of from January we use the one-month US Treasury bill rate.

As a proxy for e w f ollowed a lternative e nergy s tock i ndices f or t he s ample p eriod f rom J anuary w idely f ollowed a lternative e nergy s tock i ndices f or t he s ample p eriod f rom J anuary v alue, b ut a s a

s et o f e xplanatory f actors t hat t he a forementioned l iterature Euronext.

T he E uropean R enewable E nergy I ndex ( ERIX), w hich i ncorporates t he 1 0 l argest a nd m ost ample e xamine b egins behaviour. f ive the w We hen tidely rue e d xamine ata f undamental 3f or a f ll ive indices become ta 4 the excess return on investments in fossil fuels, we employ ve w idely f ollowed a lternative e nergy s tock i ndices f or t he s ample p eriod f rom J anuary 04 t o J uly 2 013.

F or e ase o f c omparability, t he s ample b egins w hen d ata f or a ll i ndices b ecome As a robustness check, we also run rolling regressions with a fixed window length of 252 trading days.

Results liquid c ompanies f rom t he w ind, s olar, b iomass a nd ater e nergy s ector, i s e mployed t o c over t he identified a s k ey d eterminants o f r enewable e nergy s tocks’ r eturn b ehaviour.

2004 t o J uly 2 013.

F or e ase o f c omparability, t he s ample b egins w hen d ata f or a ll i ndices b ecome dollars. Unless otherwise stated, we collect data examine five lternative edwnergy sptock or the sample pthe eriod from January are very compared tofaollowed the filter approach.

Goldman Sachs Commodity Index (GSCI) Energy 013.

For similar ase f cwomparability, he European saample hen data for ian aind ll Utherwise indices become ailable.

Tehe toime sidely eries dKalman enominated n re S ollars.

U nless ondices stated, wae dS&P ata mUarket.

Both rice total rfeturn ata re pollect rovided bstated, y the index roprietor Société available.

Tre he ttime series baiegins denominated S dollars.

Udnless octherwise we cpollect data time s eries a re d enominated i n U S d ollars.

U nless o therwise s tated, w e c ollect d ata Total Return Index.

This benchmark index provides 4 to July 2013.

For D ease of comparability, tinternational he sample sbtock egins when ata ftor aSll indices blean ecome Générale.

O ur i ndices c omprise he &P G lobal C E nergy I ndex, t he WilderHill om Thomson Rfrom euters atastream.

Thomson Reuters Datastream. an Energy which tracks performance oif n US dollars. Unless otherwise stated, we collect data Reuters DIndex, atastream. lable.

The tData ime series are the denominated 7 idely followed alean lternative energy stock indices for the sample period r the US mIarket, e focus n w the Energy Index, which tracks the performance of from January exchanges. n JWe uly e2wxamine 013, the five iondex consisted of 5C1 4WilderHill Ase awe robustness check, we also run rolling regressions withwahich fixedtracks windowthe length of 252 tradingodays.

For UilderHill S m arket, w focus on run the WilderHill Clean Ewith nergy Index, performance f Results 3 Ro As athe robustness also regressions a fixed m Thomson euters atastream. rket, we focus n the WD Ccheck, lean Energy Index, wrolling hich tracks the performance of k

k

k

k

k

k

k

k

k

k

∏

2004 to JSuly 2013. or ajor eare ase oS f sctock omparability, the In sample begins wapproach. hen data for all indices ecome ly the window ew York tock Exchange (NYSE) an from energy cNompanies listed o252 n Fm U exchanges.

July 2013, the index consisted of 51 bmakes very similar Kalman filter length of trading Results areUto very similar compared 4 French the data oavailable clean nergy ompanies ldays. isted n compared m2ajor S the stock xchanges. n toJuly 2013, the index consisted f 51 at http://mba.tuck.dartmouth.edu/ ompanies lbtain isted omn onthly meajor U S sctock exchanges.

Iost n oJuly 013, tUhe ceonsisted o f 5I1 Tfilter he time slargest eries aata re ddenominated in the S N dindex ollars.

Unless otherwise stated, we collect data (ERIX), we hich iavailable. ncorporates total he 10 and mirectly ocks.

W o t r eturn d f rom ew Y ork S tock E xchange ( NYSE) the Kalman approach. pages/faculty/ken.french/data_library.html. the U S m arket, w e f ocus o n t he W ilderHill C lean E nergy I ndex, w hich t racks t he p erformance o f stocks.

W obtain m onthly the otal return ata Edxchange irectly f(rom the New York Stock Exchange (NYSE) ain monthly total rTeturn dee mployed ata irectly fcatastream. rom New York Sd tock NYSE) homson Rd euters D d water nergy sector, iRs enewable tEo nergy over Itndex the ronext.

Tehe Efrom uropean (ERIX), which incorporates the 10 largest and most n e nergy c ompanies l isted o n m ajor U S s tock e xchanges.

I n J uly 2 013, the index consisted 51 and most Euronext.

T he E uropean R enewable E nergy I ndex ( ERIX), w hich the 10 olf argest European R enewable E nergy I ndex ( ERIX), w hich i ncorporates t he 1 0 l argest a nd most 7incorporates a uid are cpompanies rovided by the index pind, roprietor Sbociété from the warket, sw olar, iomass and water eCnergy sector, Index, is employed to cover tphe For tm he U S FOR m e focus on the W ilderHill lean Energy w hich tracks the is erformance f cover the 8 • CENTRE INTERNATIONAL GOVERNANCE INNOVATION es f rom t he w ind, s olar, b iomass a nd w ater e nergy ector, i s e mployed t o c over t he cks.

W e o btain onthly t otal r eturn d ata d irectly f rom t he N ew Y ork S tock E xchange ( NYSE) liquid c ompanies f rom t he w ind, s olar, b iomass a nd w ater e nergy s ector, e mployed too e the S&P lobal CBlean Index, the WilderHill ropean mGarket. oth pEnergy rice and total risted eturn data are UpS rovided by the index proprietor Société clean e nergy c ompanies l o n m ajor s tock e xchanges.

I n J uly 2 013, t he i ndex c onsisted o f 5 1 ket. Both price and total rReturn data re pprovided by ttotal he index pwroprietor Société European m Baoth rice aIndex nd return data are provided bthe y tthe he index proprietor Société onext. uropean enewable Energy (SERIX), hich incorporates 10 largest and most nérale. TOhe ur E international sarket. tock omprise the &P Global Clean EN nergy Index, W ilderHill stocks. We obtain mindices onthly tcotal return data directly from the ew York Stock Exchange (NYSE)

international stock indices comprise the S&P Global Clean Energy Index, the WilderHill

Générale. ur international stock aindices comprise the S&P Gis lobal Clean Etnergy Index, s with a fixed window length 252 trading Results id companies from the Oof w ind, solar, days. biomass nd water energy sector, employed o cover the the WilderHill

WHAT DROVE THE MID-2000s’ EXPLOSIVENESS IN ALTERNATIVE ENERGY STOCK PRICES? EVIDENCE FROM US, EUROPEAN AND GLOBAL INDICES Table 1: Summary Statistics Mean Return (in %)

Minimum (in %)

Maximum (in %)

Standard Deviation (in %)

Skewness

Kurtosis

WilderHill Clean Energy Index

–0.26 (–0.25)

–32.06

26.72

9.42

–0.42

3.87

ERIX

0.79 (0.65)

–40.82

33.45

11.03

–0.51

4.70

S&P Global Clean Energy Index

0.09 (0.08)

–39.94

22.56

9.57

–1.15

5.81

WilderHill New Energy Global Innovation Index

0.50 (0.53)

–35.00

21.63

8.29

–0.96

5.74

Ardour Global Alternative Energy Index Composite

0.32 (0.32)

–38.72

23.92

8.96

–1.06

6.13

Index

Note: This table reports summary statistics for monthly total return alternative energy stock indices. The sample period runs from January 2004 to July 2013. Heteroskedasticity and autocorrelation consistent standard errors (Newey and West 1987) are used to compute the t-statistics reported in parentheses.

exposure to six energy commodities including crude oil, Brent crude oil, unleaded gasoline, heating oil, gas oil and natural gas. The index tracks not only the prices of the underlying first nearby futures contracts (spot yield), but also factors in the discount or premium obtained by rolling forward hypothetical positions in such contracts as they approach delivery (roll yield) as well as the interest earned on hypothetical fully collateralized contract positions (collateral yield).

For the bubble detection tests, we use daily price data as this frequency enables more powerful tests. We deflate the daily price time series in order to obtain real index prices.

To this end, we use the following consumer price indices (CPI): CPI All Urban Consumers from the US Bureau of Labor Statistics, the Euro Area HICP All Items from Eurostat and the OECD-Total CPI All Items. As the CPIs are only available at monthly frequency, they are linearly interpolated to obtain daily time series (see, for example, Homm and Breitung 2012).

To construct sector-specific indices, the stock price indices are orthogonalized with respect to the return variations of a leading stock market index, a technology stock index and the WTI crude oil spot price.5 For the broad stock market portfolios, we use the US price index S&P 500, the Morgan Stanley Commodity Index (MSCI) Europe Investable Market Indices (IMI) and the MSCI World IMI, respectively.

Note that the MSCI World IMI also contain small-cap stocks and thus have broader market coverage than the standard index versions. To control for the exposure to technology stocks, we employ the NYSE Arca Tech 100 for the United States, the STOXX Europe TMI Technology and the MSCI World IMI Information Technology, respectively.

5 Our results are robust to the use of the S&P GSCI Energy Spot Index that tracks the prices of the futures contracts of six different fossil fuels.

The index is highly correlated with WTI (West Texas Intermediate, or Texas light sweet) crude oil (Pearson’s r = 0.89).

EMPIRICAL RESULTS PERFORMANCE ANALYSIS To obtain a first impression of the return behaviour of our alternative energy stock indices, we report descriptive statistics in Table 1. Note that none of the indices is able to produce a significantly positive total return over the sample period between January 2004 and July 2013. The monthly mean raw return of the US WilderHill Clean Energy Index is even slightly negative. The minimum and maximum return statistics are quite extreme and a result of the relatively high monthly volatility. Moreover, the return distributions exhibit negative skewness and elevated kurtosis.

Table 2 reports that US, European and global renewable energy stock indices deliver negative monthly riskadjusted returns over the entire sample period. For the WilderHill Clean Energy Index and the S&P Global Clean Energy Index, the negative abnormal returns are quite substantial and even statistically significant. Given their high market betas and the mostly significant loadings on the size factor, the indices not only turn out to be lowreturn, but also rather risky investments. Except for the ERIX, the indices are somewhat tilted toward the growth segment. Moreover, the loadings on price momentum are insignificant over the entire sample period.

Controlling for exposure to fossil fuels yields an interesting picture. First, the five-factor model alphas and the factor loadings in Panel B of Table 2 do not change much compared to their four-factor counterparts in Panel A. Second, only the US index has a significantly positive sensitivity to price variations in the oil futures market. For the international index S&P Global Clean Energy Index, the coefficient is significant at the 10 percent level, but quite low in economic terms. This finding suggests some subtle differences in the perception of renewable energies Martin T. Bohl, Philipp Kaufmann and Pierre L. Siklos • 9

CIGI Papers no. 36 — July 2014 Table 2: Multifactor Model Performance Index

Alpha (in %)

RMRF

SMB

HML

WML

Energy

Adj. R2

WilderHill Clean Energy Index

–1.38** (-2.36)

1.60*** (12.66)

0.85*** (3.09)

–0.43* (–1.92)

0.03 (0.25)

0.67

ERIX

–0.62 (–0.83)

1.40*** (9.99)

1.16*** (3.64)

0.52 (1.20)

0.27 (1.51)

0.63

S&P Global Clean Energy Index

–1.08* (–1.76)

1.68*** (14.47)

0.44 (1.19)

–0.08 (–0.19)

0.13 (0.83)

0.68

WilderHill New Energy Global Innovation Index

–0.49 (–1.13)

1.51*** (17.80)

0.67** (2.54)

–0.23 (–0.84)

–0.04 (–0.42)

0.78

Ardour Global Alternative Energy Index Composite

–0.80 (–1.48)

1.60*** (14.10)

0.66** (2.05)

–0.12 (–0.34)

0.10 (0.65)

0.73

WilderHill Clean Energy Index

Panel A: Four-Factor Model

Panel B: Five-Factor Model –1.31 (–2.37)

***

1.34 (10.03)

1.00*** (3.87)

–0.52** (–2.50)

–0.08 (–0.80)

0.23*** (3.72)

0.71

ERIX

–0.57 (–0.82)

1.31*** (7.31)

1.12*** (3.31)

0.60 (1.41)

0.24 (1.40)

0.08 (1.05)

0.63

S&P Global Clean Energy Index

–1.04* (–1.72)

1.59*** (13.74)

0.40 (1.10)

–0.03 (–0.09)

0.09 (0.61)

0.08* (1.87)

0.68

WilderHill New Energy Global Innovation Index

–0.46 (–1.06)

1.45*** (15.92)

0.65** (2.30)

–0.20 (–0.73)

–0.07 (–0.66)

0.06 (1.58)

0.79

Ardour Global Alternative Energy Index Composite

–0.78 (–1.43)

1.54*** (12.12)

0.64* (1.94)

–0.09 (–0.25)

0.08 (0.52)

0.05 (1.07)

0.73

**

Note: This table reports results for the Carhart (1997) four-factor model and the five-factor model of equations (1) and (2), respectively. The regressions are performed using monthly total return data. The sample period runs from January 2004 to July 2013. Heteroskedasticity and autocorrelation consistent standard errors (Newey and West 1987) are used to compute the t-statistics reported in parentheses. *, ** and *** denote statistical significance at the 10, five and one percent levels, respectively.

across continents. In the United States, investors appear to view alternative energy stocks as a hedge against rising crude oil prices (Sadorsky 2012a) and could thus be more likely to adopt a cost-benefit attitude toward the further deployment of renewable energy technologies. In Europe, however, the promotion of renewable energies is mostly independent of the price developments in oil markets.

Instead, it is viewed not only as a means of achieving climate targets or replacing nuclear power in the long run, but also as a device to increase public awareness of a sustainable energy supply.

The subperiod regressions in Table 3 split the sample into two periods from January 2004 to December 2007 and from January 2008 to July 2013, respectively.

Some insightful patterns emerge in comparison to the full sample results. All sector indices exhibit a severe deterioration in abnormal performance from the first to the second subperiod. Since 2008, the S&P Global Clean Energy Index, for instance, loses 2.58 percent per month on average after adjusting for factor risks. From 2004 to 2007, however, the index earns a positive though insignificant five-factor abnormal return. This pronounced reversal from slightly positive to significantly negative alphas is also mirrored in the coefficients on price momentum. A large portion of the strikingly positive performance until December 2007 can be explained by momentum. With the exception of the US index, all indices load significantly on 10 • CENTRE FOR INTERNATIONAL GOVERNANCE INNOVATION

the WML factor during the first subperiod, indicating that renewable energy stocks belonged to the group of winner stocks in the cross-section of the market. However, the momentum exposure completely disappears in the second subperiod and becomes even significantly negative for the US index. As discussed previously, positive payoffs from an investment in fossil fuels only have a positive impact on US alternative energy stocks. The sensitivity becomes insignificant after 2007, suggesting a structural break in the relationship.

With regard to the other benchmarks, the indices are somewhat more sensitive to market fluctuations and mostly have a stronger small-cap tilt in the second subperiod. Note that the systematic factors explain the return variations of our indices fairly well. For the subperiod regressions, the adjusted R2 attains values of up to 0.81. By contrast, Bohl, Kaufmann and Stephan (2013) show that the explanatory power for their German sample stocks often does not exceed the value of 0.5, which indicates a larger idiosyncratic component.

IDIOSYNCRATIC PRICE TIME SERIES AND FACTOR EXPOSURES Having analyzed the performance drivers of our international renewable energy stock sample, we now consider whether the indices contained a speculative

WHAT DROVE THE MID-2000s’ EXPLOSIVENESS IN ALTERNATIVE ENERGY STOCK PRICES? EVIDENCE FROM US, EUROPEAN AND GLOBAL INDICES Table 3: Subperiod Performance Analysis Index

Alpha1

Alpha2

RMRF1

RMRF2

SMB1

SMB2

HML1

HML2

WML1

WML2

Energy1

Energy2

Adj. R2

WilderHill Clean Energy Index

0.79% (1.54)

–2.89%*** (–4.45)

0.75*** (2.62)

1.54*** (11.11)

1.52*** (3.67)

0.77** (2.61)

–0.69** (–2.05)

–0.69*** (–3.52)

0.55 (1.48)

–0.18** (–2.59)

0.21** (2.27)

0.10 (1.39)

0.76

Difference European Renewable Energy Index Difference S&P Global Clean Energy Index Difference WilderHill New Energy Global Innovation Index Difference Ardour Global Alternative Energy Index Composite Difference

–3.67%*** (–4.19) 0.19% (0.23)

–1.92%* (–1.70)

–2.11% (–1.51) 0.58% (1.06)

–2.58%*** (–3.13)

–3.16%*** (–3.13) 0.61% (1.20)

–1.48%** (–2.59)

–2.09%*** (–2.65) –0.20% (–0.24)

–1.72%** (–2.44)

–1.52% (–1.33)

+0.79** (2.15) 1.13*** (3.40)

1.29*** (4.73)

+0.17 (0.38) 1.23*** (5.39)

1.65*** (9.67)

+0.43 (1.35) 1.02*** (5.52)

1.49*** (10.22)

+0.47* (1.85) 1.43*** (3.89)

1.54*** (8.06)

+0.11 (0.24)

–0.75 (–1.41) 0.73* (1.74)

1.18*** (3.78)

+0.45 (0.87) –0.12 (–0.35)

0.61 (1.20)

+0.73 (1.19) 0.42 (1.56)

0.69* (1.93)

+0.27 (0.57) –0.16 (–0.37)

0.93** (2.54)

+1.10* (1.95)

0.00 (0.00) –0.73 (–0.82)

0.39 (0.69)

+1.13 (1.04) –0.27 (–0.41)

–0.27 (–0.64)

0.00 (0.00) –0.38 (–0.97)

–0.38 (–1.17)

0.00 (0.00) –0.61 (–0.91)

–0.11 (–0.27)

+0.50 (0.62)

–0.73** (-2.15) 2.14*** (3.26)

0.01 (0.06)

–2.13*** (–3.16) 1.15*** (3.21)

–0.03 (–0.22)

–1.18*** (–3.28) 0.90*** (3.41)

–0.17 (–1.64)

–1.08*** (–3.84) 1.28** (2.35)

–0.05 (–0.34)

–1.32*** (–2.77)

–0.11 (–1.03) 0.09 (1.18)

–0.01 (–0.06)

0.68

–0.10 (–0.51) 0.06 (0.96)

–0.02 (–0.15)

0.72

–0.07 (–0.70) 0.00 (0.04)

0.00 (–0.02)

0.81

0.00 (–0.04) 0.05 (0.62)

–0.02 (–0.22)

0.76

–0.07 (–0.67)

Note: This table reports results for the dummy variable five-factor model regression of equation (3). The binary dummy is equal to zero from January 2004 to December 2007 and equal to one from January 2008 to July 2013. This approach yields monthly estimates for two subperiods indicated by a subscript. The corresponding differences between subperiods are presented below. Heteroskedasticity and autocorrelation consistent standard errors (Newey and West 1987) are used to compute the t-statistics reported in parentheses. *, ** and *** denote statistical significance at the 10, five and one percent levels, respectively.

bubble in the mid-2000s. Bohl, Kaufmann and Stephan (2013) are the first to tackle this issue and conclude that German alternative energy stocks exhibited explosive price behaviour. However, they remain silent on potential explanations for the explosiveness and merely associate the price surge between 2005 and 2007 with exuberant investor sentiment toward the promising energy sector.

We extend this line of research by examining factors that may explain the bubble-like phenomenon.

Figure 1 shows the performance of both the real price indices and the real idiosyncratic price indices. The latter reflect the sector-specific performance after removing the effects of general market fluctuations, technology stocks and oil prices. The graphs also reveal the weaker performance of US alternative energy stocks compared to their European and international counterparts — a result that we have already encountered in the previous section.

Overall, investor sentiment toward the clean energy sector seems to have been far less exuberant in the United States than in Europe. Consistent with the mean raw return reported in Table 1, the ERIX exhibits the sharpest rise in prices prior to 2008.

The sector-specific components of our European and global indices gradually increase and do not start to decline until after the world stock markets have begun their recovery in early 2009. Except for the US sample, the peaks in the idiosyncratic price indices are all located in May 2009. This suggests that the severe crashes following the outbreak of the global financial crisis in 2008 were mainly driven by systematic risk factors, such as the relatively high market betas. By contrast, the US idiosyncratic price index steadily declines throughout the sample period. This is an indication that the positive performance until the end of 2007 can largely be attributed to our set of exogenous factors, whereas the sector-specific component was mostly negative.

Figures 2–6 display the alpha coefficients as well as the time-varying exposures of the multifactor model in equation (8). The factor loadings are computed as the sum of the contemporaneous and lagged coefficients (see Dimson 1979). The daily alphas somewhat reflect the aforementioned reversal in abnormal returns after 2008.

However, they should not be directly compared with the estimates in Tables 2 and 3, as the latter are based on Martin T. Bohl, Philipp Kaufmann and Pierre L. Siklos • 11

CIGI Papers no. 36 — July 2014 Figure 1: Real Price Indices and Real Idiosyncratic Price Indices WilderHill Clean Energy Index

European Renewable Energy Index

200

700 600

150

500 400

100

300

200

50

100 0 2004

2005

2006

2007

2008

2009

ECO

2010

2011

2012

2013

0 2004

2005

ECO Idiosyncratic

2006

2007

2008

2009

ERIX

S&P Global Clean Energy Index

2010

2011

2012

2013

ERIX Idiosyncratic

WilderHill New Energy Global Innovation Index

400

300 250

300 200

200

150 100

100 50 0 2004

2005

2006

2007

2008

2009

SPGlobal

2010

2011

2012

2013

SPGlobal Idiosyncratic

0 2004

2005

2006

NEX

2007

2008

2009

2010

2011

2012

2013

NEX Idiosyncratic

Ardour Global Alternative Energy Index Composite 300 250 200

150 100 50 0 2004

2005

2006

2007

Ardour

2008

2009

2010

2011

2012

2013

Ardour Idiosyncratic

This figure shows the inflation-adjusted stock price indices and the inflation-adjusted, sector-specific price indices. The latter represent the idiosyncratic portion after controlling for time-varying exposures to a set of explanatory variables. Both indices are based on daily data for the sample period from January 2004 to July 2013.

monthly excess returns and on a different set of benchmark factors. For all indices, the sensitivity to the broad market index is relatively strong, corroborating our earlier findings for the monthly performance evaluation. For example, the daily market beta of the S&P Global Clean Energy Index fluctuates quite heavily between 0.59 and 2.71, averaging 12 • CENTRE FOR INTERNATIONAL GOVERNANCE INNOVATION

1.49. Interestingly, the correlation with technology stocks is strongest for the US index, while the European and global indices are less exposed to return variations in the tech sector. The loading of the WilderHill Clean Energy Index on the NYSE Arca Tech 100 ranges between 0.44 and 1.55, whereas the exposure of the European Renewable Energy

WHAT DROVE THE MID-2000s’ EXPLOSIVENESS IN ALTERNATIVE ENERGY STOCK PRICES? EVIDENCE FROM US, EUROPEAN AND GLOBAL INDICES Figure 2: Time-varying Factor Exposures of the WilderHill Clean Energy Index Alpha

S&P 500

0.4%

2.0 1.5

0.2% 1.0 0.0%

0.5

0.0 -0.2% -0.5 -0.4% 2004

2005

2006

2007

2008

2009

2010

2011

2012

2013

-1.0 2004

2005

2006

NYSE Arca Tech 100

2007

2008

2009

2010

2011

2012

2013

2010

2011

2012

2013

WTI Crude Oil

2.0

0.6 0.4

1.5

0.2 1.0

0.0 0.5

0.0 2004

-0.2

2005

2006

2007

2008

2009

2010

2011

2012

2013

-0.4 2004

2005

2006

2007

2008

2009

shows daily time-varying factor exposures for the WilderHill Clean Energy Index. The Kalman filter is used to estimate the time-varying This figure factor loadings of equations (8) to (10) for the period from January 2004 to July 2013. Except for the alpha, the solid lines represent the sum of the contemporaneous and lagged coefficients, whereas the dotted lines represent confidence bands for the 10 percent level of statistical significance.

Index to the STOXX Europe TMI Technology lies only in a narrow and mostly insignificant range between –0.04 and 0.18. For the global indices, the sensitivity to technology stocks sometimes falls into deeper negative territory.

As far as the sensitivity to oil price fluctuations is concerned, we find an interesting pattern. Prior to the burst of the oil price bubble in mid-2008, all indices show a positive and mostly significant reaction to an increase in crude oil prices. For the WilderHill Clean Energy Index, the exposure to oil price changes ranges between 0.09 and 0.33 during that period of time. However, the positive relationship breaks down over the course of the bubble crash in October 2008. For the European and global indices, the oil sensitivity is lower prior to 2008, but still of some economic relevance. In summary, we can conclude that rising oil prices partly drove the performance of alternative energy stocks across the globe, with the effect being most pronounced for the US sector. However, in the aftermath of the global economic crisis and the bursting of the 2008 oil bubble, the positive relationship diminishes or even disappears entirely.

Note that prior to mid-2008, the daily exposures to WTI crude oil are somewhat higher than the average monthly loadings on the S&P GSCI Energy Total Return Index reported in Tables 2 and 3. These two assets do not necessarily move in tandem, as the latter total return index also captures gains or losses associated with rolling forward futures contracts once they reach their expiry dates. This so-called roll yield is positive if the futures market is in backwardation, meaning that the expected spot price exceeds the futures prices. The roll yield is negative if the futures market is in contango, a situation where the futures curve is upward sloping. For instance, after the oil bubble burst in July 2008, the oil futures market was quickly entering a sustained period of contango (see, for example, Kesicki 2010). As a result, the payoff from the S&P GSCI Energy Total Return Index was rather moderate, although spot prices of fossil fuels began to rise again.

SADF TESTS Visual inspection of the charts in Figure 1 indicates some explosive price behaviour, at least for the European and Martin T. Bohl, Philipp Kaufmann and Pierre L. Siklos • 13

CIGI Papers no. 36 — July 2014 Figure 3: Time-varying Factor Exposures of the European Renewable Energy Index Alpha

MSCI Europe IMI 2.5

0.6% 0.4%

2.0

0.2%

1.5

0.0%

1.0

-0.2%

0.5

-0.4% -0.6% 2004

2005

2006

2007

2008

2009

2010

2011

2012

2013

0.0 2004

2005

2006

2007

STOXX Europe TMI Technology 0.3

0.2

0.2

0.1

0.1

0.0

0.0

-0.1

-0.1

2005

2006

2007

2008

2009

2010

2011

2009

2010

2011

2012

2013

2010

2011

2012

2013

WTI Crude Oil

0.3

-0.2 2004

2008

2012

2013

-0.2 2004

2005

2006

2007

2008

2009

This figure shows daily time-varying factor exposures for the European Renewable Energy Index. The Kalman filter is used to estimate the timevarying factor loadings of equations (8) to (10) for the period from January 2004 to July 2013. Except for the alpha, the solid lines represent the sum of the contemporaneous and lagged coefficients, whereas the dotted lines represent confidence bands for the 10 percent level of statistical significance.

Figure 4: Time-varying Factor Exposures of the S&P Global Clean Energy Index Alpha

MSCI World IMI

0.4%

4.0

0.2%

3.0

0.0%

2.0

-0.2%

1.0

-0.4%

0.0

-0.6% 2004

2005

2006

2007

2008

2009

2010

2011

2012

2013

-1.0 2004

2005

2006

MSCI World IMI Information Technology

2007

2008

2009

2010

2011

2012

2013

2010

2011

2012

2013

WTI Crude Oil

2.0

0.3 0.2

1.0

0.1 0.0

0.0 -1.0

-2.0 2004

-0.1

2005

2006

2007

2008

2009

2010

2011

2012

2013

-0.2 2004

2005

2006

2007

2008

2009

This figure shows daily time-varying factor exposures for the S&P Global Clean Energy Index. The Kalman filter is used to estimate the time-varying factor loadings of equations (8) to (10) for the period from January 2004 to July 2013. Except for the alpha, the solid lines represent the sum of the contemporaneous and lagged coefficients, whereas the dotted lines represent confidence bands for the 10 percent level of statistical significance.

14 • CENTRE FOR INTERNATIONAL GOVERNANCE INNOVATION

WHAT DROVE THE MID-2000s’ EXPLOSIVENESS IN ALTERNATIVE ENERGY STOCK PRICES? EVIDENCE FROM US, EUROPEAN AND GLOBAL INDICES Figure 5: Time-varying Factor Exposures of the WilderHill New Energy Global Innovation Index Alpha

MSCI World IMI

0.4%

3.0

0.2% 2.0 0.0%

1.0 -0.2%

-0.4% 2004

2005

2006

2007

2008

2009

2010

2011

2012

2013

0.0 2004

2005

2006

MSCI World IMI Information Technology

2007

2008

2009

2010

2011

2012

2013

2010

2011

2012

2013

WTI Crude Oil

0.6

0.2

0.3 0.1 0.0

0.0 -0.3

-0.6 2004

2005

2006

2007

2008

2009

2010

2011

2012

2013

-0.1 2004

2005

2006

2007

2008

2009

This figure shows daily time-varying factor exposures for the WilderHill New Energy Global Innovation Index. The Kalman filter is used to estimate the time-varying factor loadings of equations (8) to (10) for the period from January 2004 to July 2013. Except for the alpha, the solid lines represent the sum of the contemporaneous and lagged coefficients, whereas the dotted lines represent confidence bands for the 10 percent level of statistical significance.

Figure 6: Time-varying Factor Exposures of the Ardour Global Alternative Energy Index Composite Alpha

MSCI World IMI

0.4%

4.0 3.0

0.2%

2.0 0.0%

1.0 -0.2%

-0.4% 2004

0.0

2005

2006

2007

2008

2009

2010

2011

2012

2013

-1.0 2004

2005

2006

MSCI World IMI Information Technology

2007

2008

2009

2010

2011

2012

2013

2010

2011

2012

2013

WTI Crude Oil

1.5

0.2

1.0 0.5

0.1

0.0

-0.5

0.0

-1.0 -1.5 2004

2005

2006

2007

2008

2009

2010

2011

2012

2013

-0.1 2004

2005

2006

2007

2008

2009

This figure shows daily time-varying factor exposures for the Ardour Global Alternative Energy Index Composite. The Kalman filter is used to estimate the time-varying factor loadings of equations (8) to (10) for the period from January 2004 to July 2013. Except for the alpha, the solid lines represent the sum of the contemporaneous and lagged coefficients, whereas the dotted lines represent confidence bands for the 10 percent level of statistical significance.

Martin T. Bohl, Philipp Kaufmann and Pierre L. Siklos • 15

CIGI Papers no. 36 — July 2014 Table 4: Bubble Detection Tests Index

Recursive SADF

Rolling SADF

GSADF

0.742

1.136

1.757

–0.026

1.352

2.382**

Panel A: Test Statistics WilderHill Clean Energy Index Price index Idiosyncratic price index

European Renewable Energy Index Price index

3.880***

3.519***

4.647***

Idiosyncratic price index

2.212***

1.368

2.660**

S&P Global Clean Energy Index Price index

3.309***

3.094***

4.417***

Idiosyncratic price index

2.329***

1.490*

3.234***

WilderHill New Energy Global Innovation Index Price index

3.074***

2.925***

3.749***

Idiosyncratic price index

2.202***

1.680*

2.334**

Ardour Global Alternative Energy Index Composite Price index

2.437***

1.450

2.910***

Idiosyncratic price index

1.665**

1.588*

2.733***

Panel B: Finite Sample Critical Values CV 90%

1.253

1.465

1.958

CV 95%

1.526

1.694

2.195

CV 99%

2.023

2.145

2.722

This table reports the test statistics of the three SADF procedures. Results are shown for both inflation-adjusted stock price indices and the deflated idiosyncratic portion of the stock price indices. For the recursive SADF and GSADF tests, the sample fraction r0, which determines the initial window length, is set to 0.1, while for the rolling SADF test the constant fraction r equals 0.2. Right-tailed critical values are obtained from Monte Carlo simulations with 5,000 replications. The sample period runs from January 2004 to July 2013. *, ** and *** denote statistical significance at the 10, five and one percent levels, respectively.

global indices. To formally test this supposition, we perform the recursive, rolling and generalized versions of the SADF test outlined in the section on “Bubble Detection Tests.” Table 4 presents the supremum test statistics as well as the associated finite sample critical values. Based on these results, we do not find any signs of explosiveness for US alternative energy stocks, which leads us to reject the possibility of a bubble. By contrast, for European and international stocks we can reject — in most cases even at the one percent level — the unit root null hypothesis in favour of the explosive alternative. The test statistics for the idiosyncratic price indices shed light on the degree to which the explosive price behaviour was driven by exogenous factors. Even after controlling for return variations in the aggregate equity market, technology stocks and crude oil, we overwhelmingly reject the null hypothesis of a unit root. This finding suggests that the explosiveness in the mid-2000s was sector-specific and not merely a result of rising oil prices or the bullish market trend in general.

However, inferences cannot be solely drawn on the basis of the supremum test statistics. The date-stamping

16 • CENTRE FOR INTERNATIONAL GOVERNANCE INNOVATION

procedure has to be taken into account as well. As an example of this, Figures 7 and 8 show the sequences of backward supremum augmented Dickey-Fuller (BSADF) test statistics and their simulated right-tailed critical values at the 95 percent significance level. Both series pertain to the GSADF test. Table 5 provides similar information and specifies the months where episodes of explosiveness occurred. These are the months when the sequences of test statistics exceeded and subsequently fell below again the critical value at the 95 percent level. Note that the datestamping procedure is only applicable if the corresponding supremum test statistic in Table 4 is statistically significant.

We only report those months that exhibit genuine explosive behaviour due to sharp rises in the real price indices and disregard explosiveness that stems from falling stock prices. Yiu, Yu and Jin (2013) refer to the latter episodes as negative bubbles. However, from a rational standpoint it remains questionable whether such a phenomenon even exists. Besides, we do not consider explosiveness in the real idiosyncratic price time series in the midst of the 20082009 global financial crisis, since the underlying real price indices had already begun to drop substantially in value at that time. The significant GSADF test statistic for the

WHAT DROVE THE MID-2000s’ EXPLOSIVENESS IN ALTERNATIVE ENERGY STOCK PRICES? EVIDENCE FROM US, EUROPEAN AND GLOBAL INDICES Table 5: Date-stamping Periods of Explosiveness Index

Recursive SADF

Rolling SADF

GSADF

Price index

n/a

n/a

n/a

Idiosyncratic price index

n/a

n/a

n/a

Price index

06/2005–05/2006 12/2006–01/2008

01/2006–05/2006 04/2007–07/2007

08/2005–10/2005 01/2006–05/2006 02/2007–11/2007

Idiosyncratic price index

06/2005–11/2005 01/2006–05/2006

n/a

08/2005–10/2005 04/2006–05/2006

WilderHill Clean Energy Index

European Renewable Energy Index

S&P Global Clean Energy Index Price index

09/2005–10/2005 01/2006–05/2006 12/2006–01/2008

01/2006–05/2006 10/2007–12/2007

09/2005–10/2005 01/2006–05/2006 04/2007–12/2007

Idiosyncratic price index

09/2005–10/2005 02/2006–06/2006 10/2007–02/2008

02/2006–05/2006 12/2007–02/2008

09/2005–10/2005 02/2006–05/2006 11/2007–01/2008

WilderHill New Energy Global Innovation Index Price index

08/2005–10/2005 01/2006–06/2006 12/2006–01/2008

01/2006–05/2006 04/2007–08/2007

01/2006–05/2006 04/2007–11/2007

Idiosyncratic price index

06/2005–10/2005 01/2006–07/2006 07/2007–02/2008

01/2006–06/2006 12/2007–01/2008

02/2006–05/2006 11/2007–12/2007

Ardour Global Alternative Energy Index Composite Price index

04/2006–05/2006 04/2007–01/2008

n/a

04/2006–05/2006 07/2007–12/2007

Idiosyncratic price index

12/2007–01/2008

12/2007–02/2008

11/2007–01/2008

This table reports the months in which episodes of explosiveness originated and subsequently terminated again. The sample period runs from January 2004 to July 2013.

idiosyncratic WilderHill Clean Energy Index is therefore also misleading because Figure 8 shows that it stems from late 2011, a time where the real price index was tumbling and thus caused negative explosiveness. The bottom line is that one has to be cautious when it comes to interpreting the results of sequential ADF-type bubble tests.

The explosive episodes reported in Table 5 mainly cover months in the second half of 2005, the first half of 2006 as well as throughout 2007 and early 2008. These periods refer to both the real price and real sector-specific price time series, and thus suggest temporary episodes of bubble-like behaviour in European and global renewable energy stock indices. However, these periods are rather short-lived because the price overreactions were relatively quickly corrected. Recall that the sector-specific indices continued to rise until May 2009, a time when global equity markets had already begun their strong recovery from the preceding bear market (see Figure 1). As a result, the crash of the alternative energy stock indices that coincided with the outbreak of the 2008-2009 global financial crisis was not

the burst of a bubble but rather attributable to the relatively high sensitivity to market fluctuations. We therefore have reason to doubt the existence of an inflated speculative bubble similar to the one occurring during the dotcom era in the late 1990s. The delayed decline of idiosyncratic price time series suggests a substantial revaluation of alternative energy stock prices in light of intensifying competition and shrinking sales margins. This reassessment in early 2009 is consistent with substantial changes in the underlying fundamentals. First, soaring fossil fuel prices were not an issue any more as they fell back to sustainable levels after the global economic crisis. This probably dampened investors’ anticipation of further immediate government support for the promotion of renewable energies as a long-term substitute. Second, growing excess supply and overcapacity in the solar and wind turbine sector have left their mark on the industry’s profitability. Average factorygate prices for crystalline silicon solar modules began to drop in the third quarter of 2008 after having reached their peak at US$4.13 per watt of generating capacity. Since

Martin T. Bohl, Philipp Kaufmann and Pierre L. Siklos • 17

CIGI Papers no. 36 — July 2014 Figure 7: Sequences of BSADF Test Statistics for Real Price Indices WilderHill Clean Energy Index

European Renewable Energy Index 6

4 3

4

2

1

2

0

0

-1 -2

-2 -3 2004

2005

2006

2007

2008

2009

2010

BSADF sequence

2011

2012

2013

-4 2004

4

3

3

2

2

1

1

0

0

-1

-1

2006

2007

2008

2009

2010

BSADF sequence

2011

2007

2008

2009

2010

2011

2012

2013

95% critical value

WilderHill New Energy Global Innovation Index

4

2005

2006

BSADF sequence

S&P Global Clean Energy Index

-2 2004

2005

95% critical value

2012

2013

95% critical value

-2 2004

2005

2006

2007

2008

BSADF sequence

2009

2010

2011

2012

2013

95% critical value

Ardour Global Alternative Energy Index Composite 4 3

2 1 0

-1 -2 2004

2005

2006

2007

2008

2009

2010

BSADF sequence

2011

2012

2013

95% critical value

This figure shows the sequences of BSADF test statistics and their simulated right-tailed critical values at the 95 percent level. Both series are part of the GSADF test, which is performed for daily inflation-adjusted price index data in the sample period from January 2004 to July 2013.

then, the decline has continued to levels well below US$1 per watt. Similarly, the Bloomberg New Energy Finance Wind Turbine Price Index started to decline from €1.21 million per megawatt for delivery in the first half of 2009 to around €0.90 million for delivery in 2013.6

CONCLUDING REMARKS 6 We are grateful to Bloomberg New Energy Finance for making available their proprietary index data.

18 • CENTRE FOR INTERNATIONAL GOVERNANCE INNOVATION

Recent evidence suggests that a wave of exuberant investor sentiment pushed up prices of alternative energy stocks in the mid-2000s. We investigate whether the sector rally could have been driven by a speculative price bubble or whether it was merely a manifestation of rising crude oil prices and the prevailing bullish market conditions.

We show that US, European and international renewable energy stocks are highly sensitive to market fluctuations and are tilted toward the small-cap and growth stock segment. Based on a five-factor regression model,

WHAT DROVE THE MID-2000s’ EXPLOSIVENESS IN ALTERNATIVE ENERGY STOCK PRICES? EVIDENCE FROM US, EUROPEAN AND GLOBAL INDICES Figure 8: Sequences of BSADF Test Statistics for Real Idiosyncratic Price Indices WilderHill Clean Energy Index

European Renewable Energy Index

4

4

3

3 2

2

1

1

0

0

-1

-1 -2 2004

-2 2005

2006

2007

2008

2009

BSADF sequence

2010

2011

2012

2013

-3 2004

4

3

3

2

2

1

1

0

0

-1

-1

-2

-2 2006

2007

2008

2009

2010

BSADF sequence

2011

2007

2008

2009

2010

2011

2012

2013

95% critical value

WilderHill New Energy Global Innovation Index

4

2005

2006

BSADF sequence

S&P Global Clean Energy Index

-3 2004

2005

95% critical value

2012

2013

95% critical value

-3 2004

2005

2006

2007

2008

BSADF sequence

2009

2010

2011

2012

2013

95% critical value

Ardour Global Alternative Energy Index Composite 4 3 2 1 0

-1 -2 -3 2004

2005

2006

2007

2008

BSADF sequence

2009

2010

2011

2012

2013

95% critical value

This figure shows the sequences of BSADF test statistics and their simulated right-tailed critical values at the 95 percent level. Both series are part of the GSADF test, which is performed for daily inflation-adjusted idiosyncratic price index data in the sample period from January 2004 to July 2013.

we uncover a substantial deterioration in abnormal performance after the outbreak of the 2008-2009 global financial crisis. Monthly alphas turned from positive to significantly negative in the period between 2008 and 2013. Interestingly, most of the positive raw performance of European and international indices during the mid2000s can be explained by price momentum. Loadings on this factor are significantly positive until the end of 2007, but become negligibly small afterwards. By contrast, US alternative energy stocks display a rather moderate return pattern and are positively correlated with fossil fuel prices, indicating a possible hedging relationship. European and

global total return indices are, however, mainly unexposed to investment payoffs from the energy futures market.

While US alternative energy stock prices do not exhibit any signs of explosiveness in the mid-2000s, we find strong evidence of such behaviour for European and international indices. We extend previous research by accounting for factors that may be able to explain the explosive price behaviour. After controlling for market beta and exposures to technology stocks and crude oil, we still detect significant explosive deviations from the random walk path for European and international stocks. However, the idiosyncratic components did not collapse simultaneously

Martin T. Bohl, Philipp Kaufmann and Pierre L. Siklos • 19

CIGI Papers no. 36 — July 2014 with their underlying price indices during the 2008 stock market crash. Instead, they continued to increase until May 2009, when the global stock markets had already begun their recovery from the preceding bear market. This finding is suggestive of a substantial reassessment of alternative energy stock prices against the background of intensified sector competition and shrinking profit margins and does not provide conclusive evidence of a speculative bubble.

The severe crash of alternative energy stock prices in 2008 is thus unlikely to result from the bursting of a bubble, but can primarily be explained by the high market betas the stocks possess. Nevertheless, some quickly collapsing price spikes between 2005 and 2007 indicate the presence of at least some irrational exuberance among investors.

WORKS CITED Andrews, D. W. K. 1993. “Tests for Parameter Instability and Structural Change With Unknown Change Point.”

Econometrica 61 (4): 821–56.

Andrews, D. W. K. and W. Ploberger. 1994. “Optimal Tests When a Nuisance Parameter is Present Only Under the Alternative.” Econometrica 62 (6): 1383–1414.

Bettendorf, T. and W. Chen. 2013. “Are There Bubbles in the Sterling-Dollar Exchange Rate? New Evidence from Sequential ADF Tests.” Economics Letters 120 (2): 350–53.

Bloomberg New Energy Finance. 2013. Global Trends in Renewable Energy Investment 2013. Frankfurt School — UNEP Collaborating Centre for Climate & Sustainable Energy Finance and Bloomberg New Energy Finance.

Frankfurt am Main, June 2013.

Bohl, M. T., P. Kaufmann and P. M. Stephan. 2013. “From Hero to Zero: Evidence of Performance Reversal and Speculative Bubbles in German Renewable Energy Stocks.” Energy Economics 37: 40–51.

Broadstock, D. C., H. Cao and D. Zhang. 2012. “Oil Shocks and Their Impact on Energy Related Stocks in China.”

Energy Economics 34 (6): 1888–95.

Campbell, J. Y. and P. Perron. 1991. “Pitfalls and Opportunities: What Macroeconomists Should Know about Unit Roots.” NBER Macroeconomics Annual 6: 141–201.

Carhart, M. M. 1997. “On Persistence in Mutual Fund Performance.” Journal of Finance 52 (1): 57–82.

Dimson, E. 1979. “Risk Measurement When Shares are Subject to Infrequent Trading.” Journal of Financial Economics 7 (2): 197–226.

Durbin, J. and S. J. Koopman. 2001. Time Series Analysis by State Space Methods. Oxford University Press, New York.

Elton, E. J., M. J. Gruber, S. Das and M. Hlavka. 1993.

“Efficiency with Costly Information: A Reinterpretation of Evidence from Managed Portfolios.” Review of Financial Studies 6 (1): 1–22.

Eyraud, L., B. Clements and A. Wane. 2013. “Green Investment: Trends and Determinants.” Energy Policy 60: 852–65.

Fama, E. F. and K. R. French. 1993. “Common Risk Factors in the Returns on Stocks and Bonds.” Journal of Financial Economics 33 (1): 3–56.

———. 2012. “Size, Value, and Momentum in International Stock Returns.” Journal of Financial Economics 105 (3): 457–72.

20 • CENTRE FOR INTERNATIONAL GOVERNANCE INNOVATION

WHAT DROVE THE MID-2000s’ EXPLOSIVENESS IN ALTERNATIVE ENERGY STOCK PRICES? EVIDENCE FROM US, EUROPEAN AND GLOBAL INDICES Gutierrez, L. 2013. “Speculative Bubbles in Agricultural Commodity Markets.” European Review of Agricultural Economics 40 (2): 217–38.

Yiu, M. S., J. Yu and L. Jin. 2013. “Detecting Bubbles in Hong Kong Residential Property Market.” Journal of Asian Economics 28: 115–24.

Henriques, I. and P. Sadorsky. 2008. “Oil Prices and the Stock Prices of Alternative Energy Companies.” Energy Economics 30 (3): 998–1010.

Homm, U. and J. Breitung. 2012. “Testing for Speculative Bubbles in Stock Markets: A Comparison of Alternative Methods.” Journal of Financial Econometrics 10 (1): 198– 231.

Kesicki, F. 2010. “The Third Oil Price Surge — What’s Different This Time?” Energy Policy 38 (3): 1596–1606.

Kim, C. J. and C. R. Nelson. 1999. State-Space Models with Regime Switching. MIT Press, Cambridge.

Kumar, S., S. Managi and A. Matsuda. 2012. “Stock Prices of Clean Energy Firms, Oil and Carbon Markets: A Vector Autoregressive Analysis.” Energy Economics 34 (1): 215–26.

Newey, W. K. and K. D. West. 1987. “A Simple, Positive Semi-Definite, Heteroskedasticity and Autocorrelation Consistent Covariance Matrix.” Econometrica 55 (3): 703–08.

Phillips, P. C. B. and J. Yu. 2011. “Dating the Timeline of Financial Bubbles during the Subprime Crisis.”

Quantitative Economics 2 (3) : 455–91.

Phillips, P. C. B., Y. Wu and J. Yu. 2011. « Explosive Behavior in the 1990s NASDAQ: When Did Exuberance Escalate Asset Values?” International Economic Review 52 (1): 201–26.

Phillips, P. C. B., S. P. Shi and J. Yu. 2013. “Testing for Multiple Bubbles: Historical Episodes of Exuberance and Collapse in the S&P 500.” Working paper, Yale University.

REN21. 2013. Renewables 2013 Global Status Report: Renewable Energy Policy Network for the 21st Century, Paris, France.

Sadorsky, P. 2012a. “Correlations and Volatility Spillovers between Oil Prices and the Stock Prices of Clean Energy and Technology Companies.” Energy Economics 34 (1): 248–55.

———. (2012b). “Modeling Renewable Energy Company Risk.” Energy Policy 40: 39–48.

Tokic, D. 2010. “The 2008 Oil Bubble: Causes and Consequences.” Energy Policy 38 (10): 6009–15.

———. 2012. “Speculation and the 2008 Oil Bubble: The DCOT Report Analysis.” Energy Policy 45: 541–50.

Martin T. Bohl, Philipp Kaufmann and Pierre L. Siklos • 21

ABOUT CIGI The Centre for International Governance Innovation is an independent, non-partisan think tank on international governance. Led by experienced practitioners and distinguished academics, CIGI supports research, forms networks, advances policy debate and generates ideas for multilateral governance improvements. Conducting an active agenda of research, events and publications, CIGI’s interdisciplinary work includes collaboration with policy, business and academic communities around the world.

CIGI’s current research programs focus on three themes: the global economy; global security & politics; and international law.

CIGI was founded in 2001 by Jim Balsillie, then co-CEO of Research In Motion (BlackBerry), and collaborates with and gratefully acknowledges support from a number of strategic partners, in particular the Government of Canada and the Government of Ontario.

Le CIGI a été fondé en 2001 par Jim Balsillie, qui était alors co-chef de la direction de Research In Motion (BlackBerry). Il collabore avec de nombreux partenaires stratégiques et exprime sa reconnaissance du soutien reçu de ceux-ci, notamment de l’appui reçu du gouvernement du Canada et de celui du gouvernement de l’Ontario.

For more information, please visit www.cigionline.org.

CIGI MASTHEAD Managing Editor, Publications

Carol Bonnett

Publications Editor

Jennifer Goyder

Publications Editor

Vivian Moser

Publications Editor

Patricia Holmes

EXECUTIVE President

Rohinton Medhora

Vice President of Programs

David Dewitt

Vice President of Public Affairs

Fred Kuntz

Vice President of Finance

Mark Menard

COMMUNICATIONS Communications Specialist

Declan Kellly

Public Affairs Coordinator

Erin Baxter

dkelly@cigionline.org (1 519 885 2444 x 7356) ebaxter@cigionline.org (1 519 885 2444 x 7265)