International Review of Financial Analysis 34 (2014) 222–234 Contents lists available at ScienceDirect International Review of Financial Analysis A microstructure analysis of the carbon finance market☆ Don Bredin a, Stuart Hyde b,⁎, Cal Muckley a a b University College Dublin, Ireland University of Manchester, United Kingdom a r t i c l e i n f o Article history: Received 18 October 2013 Received in revised form 10 February 2014 Accepted 12 March 2014 Available online 21 March 2014 JEL classification: G13 G14 G19 a b s t r a c t The European Union Emissions Trading Scheme is the key policy instrument of the European Commission's Climate Change Program aimed at reducing greenhouse gas emissions to 8% below 1990 levels by 2012. The key asset traded under the scheme is the European Union allowance (EUA). This article examines ultra high frequency data to assess the extent of the development in the futures market of the EU Emissions Trading Scheme. Our results indicate significant developments consistent with sequential information arrival. They also indicate a negative contemporaneous relationship between volume and volatility for all contracts. The implication is that liquidity traders dominate any role played by informed traders. Incorporating the duration between trades in our analysis has significant impact suggesting that any empirical investigation of the intra-day volume–volatility relationship needs to actively account for the impact of time elapse between trades. © 2014 Elsevier Inc. All rights reserved. Keywords: CO2 emission allowances Futures Emission trading Energy Kyoto Protocol Market microstructure 1. Introduction In January 2005 the European Union (EU) introduced formally the emission trading scheme (ETS). The scheme forms a central part of the EU agreement to cut worldwide emissions of carbon dioxide (CO2).1 Under the Kyoto agreement, the EU is committed to reduce greenhouse gas (GHG) emissions by 8% (relative to 1990 levels) by 2012.2 The aim ☆ The authors would like to thank Michael Brennan, John McConnell, Maureen O'Hara and Matthew Spiegel for valuable comments and participants at the 2009 Global Finance Academy Conference at University College Dublin, participants at the 2009 Finsia/MCFS Conference, Melbourne, the International Association for Energy Economists annual meeting 2012 in Venice and at the European Association of Environmental and Resource Economists annual meeting 2012 in Prague. This work has emanated from research conducted with the financial support of the Science Foundation Ireland under Grant Number 08/SRC/FM1389. Ike Johnson and Robert Bradley provided research assistance. ⁎ Corresponding author at: The University of Manchester, Manchester Business School, United Kingdom. E-mail addresses: don.bredin@ucd.ie (D. Bredin), stuart.hyde@mbs.ac.uk (S. Hyde), cal.muckley@ucd.ie (C. Muckley). 1 Member states of the European Union ratified the Kyoto agreement in May, 2002. The agreement became legally binding on February 16, 2005. 2 Incidentally, the European Environmental Agency (October, 2012) has indicated that the European Union as a whole will over-deliver on its Kyoto target. Ellerman and Buchner (2006) find that in 2005 data from the Pilot Phase of the ETS, the scheme has been responsible for reducing CO2 emissions by between 50 and 200 million tonnes. http://dx.doi.org/10.1016/j.irfa.2014.03.003 1057-5219/© 2014 Elsevier Inc. All rights reserved. of the EU to reduce emissions by 2012 through the EU ETS is to be achieved via two phases. The first Phase (Pilot) of trading is 2005–2007 and the second one (Kyoto), which coincides with the first compliance period of the Kyoto Protocol, is 2008–2012. After these two initial phases, the third European trading Phase will commence in 2013. The emission trading scheme issues a restricted amount of emission allowances to firms on an annual basis. At the end of each year firms must hold the required amount of emission permits to meet their emissions of CO2 over that year.3 The ETS facilitates trades for firms that are either long or short on emission permits. Non-compliance with the commitments will result in a penalty of € 40 (€ 100) per tonne of CO2 produced during the first (second) commitment period. The aim of the ETS is to generate a price signal that will encourage firms to reduce their emissions. Paolella and Taschini (2008) emphasize that the ultimate aim of this scheme (as well as the US CAAA-Title IV scheme) must be to create an environment where there is scarcity of allowances which will lead to an upward trend in prices. There has been a 3 Firms are required to submit a report verifying their emissions in any year by March 31st of the following year. D. Bredin et al. / International Review of Financial Analysis 34 (2014) 222–234 considerable amount of uncertainty associated with the price of CO2 emissions over its short life to date.4 This paper examines the performance of the EU ETS from a market microstructure perspective with the aim of understanding the information assimilation process, liquidity and the degree of market efficiency in the ETS5 during both the Pilot and the Kyoto Phase.6 To commence our study, we examine the relevance of the mixture of distributions hypothesis (MDH) and the sequential information arrival hypothesis (SIAH) in the emission trading scheme. The MDH asserts a joint dependence of volume and volatility on the underlying information flow (see, inter alia, Andersen, 1996; Clark, 1973; Tauchen & Pitts, 1983). Alternatively, while the MDH does not allow for serial dependence between volume and volatility — it is a contemporaneous relationship, the sequential information arrival hypothesis (SIAH) (Copeland, 1976) posits that traders both receive information and act on it in a sequential manner, hence a new equilibrium is not established instantaneously and the potential for a lead-lag relationship between volume and volatility exists. Yet while microstructure models (see Diamond & Verrecchia, 1987; Easley & O'Hara, 1992) suggest trade arrival times convey information neither the MDH nor the SIAH hypotheses explicitly account for the effect of time duration, the elapsed time between consecutive trades, in measuring the information content of trades. In this paper, our key contribution is to take account of duration effects on the information assimilation process in the EU ETS. The microstructure literature acknowledges that trading activity in asset markets is induced by both revisions to investor expectations of asset prices, following the arrival of new information, and by the requirements of investors for liquidity (Blume, Easley, & O'Hara, 1994; O'Hara, 1987). Recent research into the analysis of trading volume as a measure and signal of investor information and liquidity needs raises important issues concerning its interaction with other key variables of the trade process, in particular asset price volatility and trade duration (Manganelli, 2005; Xu, Chen, & Wu, 2006). However, prior empirical research customarily focuses upon the price impact of one or two specific trade process variables in isolation, rather than their combined effects. In order to understand the full dynamic interactions among the three trade process variables they should be modeled simultaneously. We adopt such an approach in this paper. We initially adopt the duration based model proposed by Xu et al. (2006) to examine the relationship between time consistent volume and volatility, while controlling for exogenous duration effects. In addition, we adopt an alternative model specification, in the vein of Manganelli (2005), which allows duration to be endogenous and to be modeled simultaneously together with 4 In April 2006, by way of an example, coincident to the unofficial release of the 2005 emission data by some of the EU member states, the EUA's price declined sharply. EU ETS spot prices had reached a high of € 30.50 prior to April 2006. Following the official release by the EU commission on the 15th May 2006, showing a larger than expected surplus in the market, the spot price fell to € 15.63 on the 17th May 2006. Given that banking EUAs was prohibited between phases, the price eventually converged to close to zero at the end of Phase 1. In Phase 1, it would appear that the cap placed on emissions was far too lax and so downward pressure on the price continued. 5 Our interpretation of these concepts draws on the work of Dacorogna et al. (2001). 6 An examination of both the Pilot as well as the Kyoto Phase is important. An understanding of the initial setting up period of the exchange may have relevant and significant implications for future exchanges or trading schemes, as well as subsequent periods of the EU ETS. For example, emission trading systems are to be established in six regions in China by 2013 and nationwide by 2015 to reduce China's emission intensity relative to its GDP. In addition, Convery and Redmond (2007) highlight that the main objective of the Pilot Phase was to get the scheme up and running and specifically that it would be fully operational by 2008, the start of Phase 2 (Kyoto Phase). As spot and futures markets, registries and monitoring, reporting and verification have been established, this principal aim of Phase 1 has been achieved. The adopted microstructure approach, in this paper, provides the first thorough indication, using volume, volatility and trade duration (time between consecutive trades), of the initial performance of the EU ETS. 223 volume and volatility. Our analysis is carried out on Phase 1 and 2 EU ETS futures and we focus on futures data provided by the ECX.7 Our key results indicate significant market developments, with reduced market frictions suggesting evidence of greater levels of efficiency present in the market. The evidence of reduced market frictions is reflected in the form of reduced durations between trades, higher volumes traded and lower volatilities. As bi-directional Granger causality is found in the trade variables, the empirical evidence is consistent with the sequential information arrival hypothesis (Copeland, 1976) rather than the predictions of the mixture of distribution hypothesis (Clark, 1973; Tauchen & Pitts, 1983). This evidence in favor of the sequential information arrival hypothesis is consistent with recent results on the EU ETS reported by Conrad, Rittler, and Rotfuß (2012) and Chevallier and Sevi (2011). Turning to the duration related information, we find that trades associated with longer durations have a higher impact on volatility. Finally, our results indicate a negative contemporaneous relationship, accounting for duration effects, between volume and volatility. The implication is that liquidity traders dominate any role played by informed traders (Easley & O'Hara, 1992). The remainder of the paper is set out as follows: Section 2 discusses the emergent literature in carbon finance with specific reference to microstructure issues. Section 3 introduces the empirical methodology, outlining the adopted approach to study the information assimilation process, liquidity and the degree of market efficiency. Section 4 presents the data and empirical results. Section 5 provides some concluding comments. 2. Recent developments in carbon finance Futures contracts were introduced on the European Climate Exchange (ECX) in March 2005, with options trading following in October 2006. Trades on the ECX are cleared through the Intercontinental Exchange (ICE), which also hosts the electronic marketplaces for the Chicago Climate Exchange. In March 2008, the ECX introduced certified emission reduction (CER) futures and options and in October 2008 it introduced an EUA-CER spread trade. There have been several studies which examine the pricing behavior and developments in both spot and futures markets. Redmond and Convery (2006) examine the EU ETS using daily data over the period 1st December 2004 to 31st July 2006 focusing on the behavior of the price of carbon in relation to energy commodities, meteorological factors and a number of other variables including dummy variables to take account of policy and regulatory issues. Mansanet Bataller et al. (2007) report that energy sources and extreme temperatures are the principal factors in the determination of CO2 price levels in 2005. Similarly, Alberola, Chevallier, and Cheze (2008) and Bredin and Muckley (2011) further document the role of fundamentals on daily spot and futures prices respectively across phases of the ETS. Chevallier (2011a) assesses the transmission of international shocks from the macroeconomy, commodities and the financial markets to the carbon market. The results show that carbon prices tend to respond negatively to an exogenous recessionary shock on economic indicators. While this latter literature examines low frequency price behavior, a relatively small number of studies investigate microstructure issues, e.g., price discovery, liquidity and intra-day volatility issues, using high frequency data. The first such study, Benz and Hengelbrock (2009) 7 Futures contracts account for 76% of the volume of trades in EUAs. Of these contracts, the vast majority are traded on the ECX which was acquired by the Intercontinental Exchange in July 2010 (Mizrach, 2010). In late April 2010, the Intercontinental Exchange (ICE) agreed to pay over US $ 600 million in cash to acquire Climate Exchange (owner of the ECX) for which it previously cleared trades. The remainder of the EUA futures contracts are largely traded on the European Energy Exchange (EEX) and Nordpool (see Mansanet-Bataller & Pardo, 2008). Both spot and futures contracts are traded on the Norwegian platform Nord Pool, which has a market share of 12.5% on the futures market. In 2010 NASDAQ QMX acquired all the shares in Nord Pool International, Nord Pool clearing ASA and Nord Pool Consulting (Mizrach, 2010). 224 D. Bredin et al. / International Review of Financial Analysis 34 (2014) 222–234 finds evidence of increased liquidity during Phase 1 for both ECX and Nord pool, with particularly strong evidence for the considerably larger ECX market. Further, these authors indicate, in terms of price impact and price discovery, that the ECX represents a price leader especially for more recently traded contracts (December 2007 and 2008). Extending this research on price discovery, Mizrach and Otsubo (2014) examine a number of issues relating to trading volumes, transaction costs (spreads), price impact and return predictability for EUA and CER contracts during 2009 (Phase 2). Consistent with Benz and Hengelbrock (2009), Mizrach and Otsubo (2014), examining Granger-Gonzalo and Hasbrouck information shares, find evidence that the ECX futures market provides 90% of the price discovery. This microstructure research establishes the primacy of the ECX platform for investigating the behavior of EUA futures.8 Another strand of the carbon microstructure emergent literature empirically examines the extent of intra-day volatility. Rotfuß (2009) provides an analysis of volatility characteristics for the first and second trading periods while Chevallier and Sevi (2011) document the superior ability of realized variance measures at forecasting volatility for EUA futures.9 Rittler (2009) shows evidence of volatility transmission from the futures market to the spot market and confirms that futures provide the leading role in price discovery. Focusing on the initial period of Phase 2, Conrad et al. (2012) demonstrate that regulatory announcements have statistically significant effects on intra-day volatility, while there is a relatively marginal influence from unexpected positive aspects of scheduled macroeconomic announcements in Germany and the United States. In this article, we extend the outlined microstructure related findings with regard to market liquidity, volumes and intra-day volatility on the EU's ETS. Specifically, we further examine the information assimilation process and the degree of market efficiency, in the ETS, during both the Pilot and Kyoto Phases of the Scheme. We examine the empirical evidence in regard to the sequential information arrival hypothesis (Copeland, 1976) and the mixture of distributions hypothesis (Clark, 1973; Tauchen & Pitts, 1983). In addition, unlike the previously discussed microstructure studies, we simultaneously model the three key trade process variables: duration (i.e. the time elapsed between consecutive trades), trade volume and asset price volatility with a view to accounting for the dynamic interactions involved and identifying the class of traders principally present in the market, whether informed or liquidity traders. Finally, we conduct a range of sensitivity tests, across model specifications, contract expirations and near run to expiration contract tests, and we compare our Pilot Phase findings with those of the Kyoto Phase. Table 1 Summary statistics for EU ETS futures. Instrument Dec. 05 Dec. 06 Dec. 07 Dec. 08 Dec. 09 Dec. 10 Dec. 11 Dec. 12 Market microstructure research must confront the problems generated by the fact that transaction data arrives at irregularly spaced intervals in time. Nevertheless, many studies simply ignore the time variation between successive trades, and analyze the data utilizing econometric techniques more appropriate for equally spaced observations. We maintain that this approach may result in a significant loss of information, since the elapsed time between consecutive trades (duration) may convey important information about the state of the market. 8 A comparison of price impacts for both EUAs and certified emission reductions (CERs) is reported by Mizrach and Otsubo (2014). The authors find that the median impact for an EUA trade is 0.0106 euro and 0.0145 for the CER trade. Mizrach and Otsubo (2014) indicate that the larger price impact of a CER trade can be explained by the larger bid-ask spreads. 9 Chevallier and Sevi (2011) also find the distribution of standardized returns to be nonnormal, contrary to the mixture of distribution hypothesis (MDH). Duration 3135 15,831 5118 126,722 248,214 261,161 301,064 262,228 Volume Mean Median Mean Median 21.63 9.99 23.15 2.32 0.59 0.70 0.69 0.38 421.76 191.00 392.00 27.47 0.05 0.05 0.05 0.01 10.32 12.07 13.22 7.04 4.78 7.57 8.25 22.99 10.00 10.00 10.00 10.00 1.00 3.00 3.00 0.76 Notes. The summary statistics are reported for the full samples of available data. The sample periods for our seven contracts are as follows: December 2005 (22nd April 2005 to 19th December 2005), December 2006 (16th June 2005 to 18th December 2006), December 2007 (3rd June 2005 to 17th December 2007) and December 2008 (28th September 2005 to 15th December 2008), December 2009 (2nd January 2009 to 14th December 2009), December 2010 (4th January 2010 to 10th December 2010), December 2011 (4th January 2011 to 19th December 2011) and December 2012 (3rd January 2012 to 17th December 2012). The duration is defined as time elapsed between consecutive trades and the mean and median figures above are measured in minutes. Two approaches have been proposed to accommodate this concern. First, Engle (2000) argues that variables in empirical microstructure models should be adjusted for duration in order to obtain timeconsistent parameter estimates. In this vein, Xu et al. (2006) incorporate a time factor by standardizing both volume and volatility with respect to elapsed time (trade duration) thereby creating time consistent measures for empirical tests. We begin our analysis, in this spirit, by utilizing a time consistent structural vector autoregression (VAR) model to characterize the dynamic relationship between our duration-based volatility and volume measures. Second, Engle and Russell (1998) and Dufour and Engle (2000) explicitly incorporate the duration between trades as an endogenous variable in a microstructure model. The inclusion of a duration variable not only addresses the estimation problems associated with irregularly occurring observations, but it also permits the information conveyed by duration to be explicitly incorporated in the modeling framework. In order to better understand the dynamic relationship between duration, volume, and volatility, we further extend the analysis by specifying a trivariate VAR framework in which we explicitly model duration. 3.2. The VAR models Initially, we formulate a bivariate VAR model where volume and volatility are standardized per unit of time, thereby accommodating the scale effect of duration on volatility and trade volume, 3. Econometric methodology 3.1. Empirical issues No. Obs zd;t ¼ μ z þ p X i¼1 vd;t ¼ μ v þ p X i¼1 azi zd;t−i þ q X ðbzi þ czi τt−i Þvd;t−i þ ut i¼0 q X avi zd;t−i þ ðbvi þ cvi τt−i Þvd;t−i þ ϵt ð1Þ i¼1 where zd,t and vd,t are duration-based volatility and volume, respectively.10 Specifically, if dt = Tt − Tt − 1 is duration, the elapsed time between h i consecutive trades at Tt and Tt − 1, then zd;t ¼ ln Zdt is the log volatility t h i per unit of time, and vd;t ¼ ln Vd t is the log volume per unit of time. t Here, we measure volatility, Zt, by the absolute value of returns, |rt| and the volume variable, Vt, by the trade size at calendar time t. The orders of the lag length for the duration-based volatility and volume variables, (p,q), are determined on the basis of the Akaike information criterion (AIC). Following Xu et al. (2006), we include an interaction 10 Both duration-based volatility and volume are per unit time and in log terms to ensure that the variables are positive. D. Bredin et al. / International Review of Financial Analysis 34 (2014) 222–234 225 Table 2 Causality tests. H0 no GC Dec. 05 Dec. 06 Dec. 07 Dec. 08 Dec. 09 Dec. 10 Dec. 11 Dec. 12 Volm. to dur. 2.528 (0.000) 3.088 (0.000) 1.496 (0.072) 0.943 (0.531) 1.657 (0.033) 0.756 (0.769) 11.767 (0.000) 9.138 (0.000) 1.753 (0.020) 1.869 (0.011) 2.526 (0.000) 3.076 (0.000) 4.309 (0.000) 4.122 (0.000) 1.745 (0.021) 1.674 (0.030) 3.132 (0.000) 2.678 (0.000) 62.923 (0.000) 88.127 (0.000) 9.964 (0.000) 2.803 (0.000) 1.831 (0.013) 9.889 (0.000) 152.19 (0.000) 158.32 (0.000) 11.27 (0.000) 1.16 (0.274) 14.84 (0.000) 9.18 (0.000) 58.05 (0.000) 9.54 (0.000) 102.19 (0.000) 2.658 (0.000) 3.99 (0.000) 44.43 (0.000) 129.74 (0.000) 214.75 (0.000) 54.73 (0.000) 14.19 (0.000) 7.77 (0.000) 7.18 (0.000) 60.63 (0.000) 256.468 (0.000) 69.034 (0.000) 7.338 (0.000) 14.758 (0.000) 4.929 (0.000) Dur. to volm. Voly. to dur. Dur. to voly. Voly. to volm. Volm. to voly. Notes. This table reports Granger causality (GC) tests on duration (dur.), volume (volm.) and volatility (voly.) derived from the tick by tick data for the December 2005, 2006, 2007, 2008, 2009, 2010 and 2011 contracts. The sample periods for our seven contracts are as follows: December 2005 (22nd April 2005 to 19th December 2005), December 2006 (16th June 2005 to 18th December 2006), December 2007 (3rd June 2005 to 17th December 2007) and December 2008 (28th September 2005 to 15th December 2008), December 2009 (2nd January 2009 to 14th December 2009), December 2010 (4th January 2010 to 10th December 2010), December 2011 (4th January 2011 to 19th December 2011) and December 2012 (3rd January 2012 to 17th December 2012). F statistics are reported, with levels of significance in brackets. term τt − i(τt = ln(dt)), that allows the volume to vary with duration in both equations.11 Given duration is informative about an asset's true value (Diamond & Verrecchia, 1987; Easley & O'Hara, 1992), it follows that duration should interact with volume to influence the characteristics of asset prices. The bivariate VAR specification provides an insightful base from which to analyze our subsequent trivariate findings, which allow for the possible exogeneity of duration effects. Specifically, the detailed bivariate VAR model, which specifies an interaction term between volume and duration in the determination of future volumes and volatility, serves to investigate, in a preliminary manner, the importance of accounting for duration effects. The shock, ut, represents the informed shock as it allows for a contemporaneous relationship between duration–volume through to volatility. Easley and O'Hara (1992) highlight that market makers will adjust their quoted prices as a result of the trading of informed traders. The shock, ϵt, represents the uninformed shock as it allows for lagged relationships. In the trivariate model, the assumption of exogenous duration variation is relaxed and duration is included as an endogenous term. This allows for interaction effects from duration to volume and to volatility and vice versa. The inclusion of both model specifications, disallowing and allowing for endogenous duration effects, is of particular interest. To capture these interactions, we extend the VAR analysis to include time as an endogenous variable. The VAR for duration, dt, volume, Vt, and volatility, Zt, is: dt ¼ μd þ p1 X γid dt−i þ q1 X i¼1 Vt ¼ μv þ Zt ¼ μ z þ p2 X γiV dt−i þ ρid V t−i þ i¼1 q2 X ρiV V t−i þ r2 X i¼0 i¼1 i¼0 q3 X r3 X γ iZ dt−i þ i¼0 ρiZ V t−i þ δid Z t−i þ εt Z t ¼ K t þ εt ð3Þ where Kt = ln k1/2 t , kt is the latent information flow and εt is the disturbance term. In the traditional MDH model, return volatility is related to the latent information flow kt concurrently, and the disturbance term is serially independent. The volatility processes, specified in Eqs. (1) and (2), generalize the MDH volatility process in Eq. (3) in several ways. The presented equations accommodate a process of dynamic adjustment or a learning effect in relation to the new information, so that return volatility is affected by possibly lagged information. For example, they permit a lagged dependence in volatility, Zt.12 If there is an absence of serial dependence, the values of the lagged coefficients in the VAR systems of equations will be insignificantly different from zero. The models simply degenerate to the traditional MDH volatility process. In contrast, significant coefficients on the lagged terms are indicative of the SIAH.13 Typically, empirical tests of the MDH model at the intraday level use transaction data aggregated up to fixed intraday intervals. Naturally, there is a loss of information in this aggregation. This loss is justified partly due to the large number of zeros, making econometric analysis relatively complex if the intervals are small. In contrast, our VAR model specifications are directly based on irregularly spaced transaction data, which is free from the problem of temporal aggregation. At the same time they account for transaction time, by the inclusion of duration. i¼1 p3 X i¼0 r1 X information and a random disturbance term. The return volatility process in the MDH framework can be written as: δiV Z t−i þ ηt ð2Þ δiZ Z t−i þ ζ t i¼1 where the lag orders are determined by the Akaike information criterion (AIC). 3.2.1. Sequential and mixed distribution information assimilation The return volatility processes, specified in Eqs. (1) and (2), are generalizations of the return process in the traditional MDH model. The MDH model postulates that return volatility depends on the flow of 11 It is then possible to examine whether duration between trades affects price adjustment to trades in the volatility equation. In the volume equation, the interaction term allows us to observe the correlation between current and past volume, while accounting for the effect of elapsed time on this relation. 3.2.2. Prevalence of informed or liquidity trading Bagehot (1974) assumes that in the market there exist traders with superior information who try to transform their informational advantage into net profits. In this framework, prices are semi-strong form efficient, they reflect all public information, but not necessarily the private information which may be in the process of being traded into prices. Informed traders are defined as traders that use the fundamentals to form reliable opinions as to whether an instrument is under or 12 This generalization is quite appropriate given that empirical evidence suggests strong volatility persistence, at the microstructure level in liquid asset markets, even after controlling for the effect of information flow. The cause of volatility persistence is usually attributed to one or a combination of microstructure imperfections including, inventory control, exchange-mandated price smoothing, and/or lagged adjustment to information among others. 13 It is worthwhile noting that the greater the extent of the sequential information arrival dynamics over time, the more slowly information is accounted for in market price. In this scenario, it is less likely that there is a prevalence of informed trading in the market. 226 D. Bredin et al. / International Review of Financial Analysis 34 (2014) 222–234 Table 3 Bivariate volume–volatility VAR (December 2005 contract). Lagged volatility Current volume Lagged volume Current dur. ∗ volume Lagged dur. ∗ volume a1 a2 a3 a4 a5 a6 a10 b0 b1 b2 b3 b4 c0 c1 Table 5 Bivariate volume–volatility VAR (December 2008 contract — full sample). Volatility equation R2 = 0.162 Volume equation R2 = 0.121 Coefficient t-Stats Coefficient t-Stats 0.206 0.051 0.044 0.074 0.052 0.072 0.037 −0.346 11.133 2.715 2.314 3.909 2.767 3.803 2.001 4.446 −0.009 −0.017 1.952 3.661 0.191 0.113 0.074 0.058 −0.111 Lagged volatility 9.858 5.750 3.755 2.933 4.247 −0.017 2.681 Notes. This table reports the results of the bivariate volume–volatility VAR. The variables are standardized with respect to duration and are in log-form to ensure that they remain positive. Estimation is based on the tick-by-tick data for the December 2006 expiry EUA futures contract traded on the ECX. The sample is 22nd April 2005 to 19th December 2005. Only coefficients statistically significant at a significance level of 5% are reported along with absolute t-statistics. Current volume Lagged volume Table 4 A bivariate volume volatility VAR summary for December 2005–07 contracts. Period b0 c0 2007 (t-value) 2006 (t-value) 2005 (t-value) −0.49 (7.64) −0.37 (9.82) −0.35 (4.45) −0.15 (7.34) −0.22 (17.32) −0.11 (4.25) Notes. This table reports the summary results of the bivariate volume–volatility VARs in respect to the Phase I contracts which expire December in each year 2005–07. In particular, it reports the contemporaneous volatility–volume coefficient, b0, and the contemporaneous duration–volume interaction coefficient, c0, in the volatility equation of these bivariate model specifications. The variables are standardized with respect to duration and are in log-form to ensure that they remain positive. Estimation is based on the tickby-tick data for EUA futures contracts traded on the ECX. The December 2005 contract is sampled 22nd April 2005 to 19th December 2005. The December 2006 contract is sampled from 16th June 2005 until 18th December 2006. The December 2007 contract is sampled from 3rd June 2005 until 17th December 2007, respectively. Only coefficients statistically significant at a significance level of 5% are reported along with absolute t-statistics. overvalued and then act on these opinions. In contrast, liquidity traders are uninformed traders that do not know the fundamental value of the instrument when forming their trading decisions (Harris, 2003). Liquidity traders are motivated by consumption needs and portfolio strategies, e.g., their trading is a reflection of their particular price sensitivities or particular trading rules (see Easley & O'Hara, 1992). Easley and O'Hara (1992) and, more recently, Manganelli (2005) highlight that although an informed trader may prefer to exploit his informational advantage by trading a larger amount of the asset, initiating this strategy would reveal his inherent informational advantage and the market maker would adjust quotes accordingly. Hence, in order to disguise his investor type, an informed trader may segment his order flow into a sequence of smaller trades separated in time.14 Consequently, given that informed traders trade to exploit their informational advantage, subsequent to the arrival of price relevant information, 14 The underlying assumption is that liquidity (noise) agents, trade with constant intensity. Current dur. ∗ volume Lagged dur. ∗ volume a1 a2 a3 a4 a5 a6 a7 a8 a9 a10 a11 a12 a13 a14 a15 a16 a17 a19 a20 b0 b1 b2 b3 b4 b5 b6 b7 b8 b9 b10 b11 b12 b13 b14 b15 b16 b17 b18 b19 b20 c0 c1 c2 c3 c4 c5 c6 c7 c8 c10 c13 c14 c15 c18 c20 Volatility equation Volume equation R2 = 0.070 R2 = 0.119 Coefficient t-Stats Coefficient t-Stats 0.169 0.052 0.031 0.023 0.020 0.016 0.011 0.017 0.009 0.012 0.008 59.869 18.332 10.797 7.899 7.067 5.446 3.932 6.101 3.214 4.027 2.722 −0.015 20.622 0.002 2.111 0.004 5.350 0.002 0.002 0.003 2.572 2.457 4.003 0.002 2.104 0.012 4.239 0.012 4.088 0.002 0.002 2.640 2.878 0.009 0.012 0.007 −0.340 0.043 0.022 0.034 0.037 3.087 4.060 2.354 31.753 3.839 1.983 3.028 3.314 0.224 0.082 0.044 0.034 0.024 0.024 0.020 0.013 0.015 0.012 0.012 0.014 0.011 0.017 0.011 0.011 0.013 0.009 0.007 0.010 79.234 28.379 14.944 11.669 8.217 8.335 6.867 4.527 5.132 4.192 4.139 4.632 3.791 5.897 3.603 3.779 4.350 3.000 2.567 3.374 −0.033 −0.033 −0.026 −0.018 −0.012 −0.011 7.511 7.457 5.910 4.024 2.748 2.530 −0.008 −0.003 6.711 2.315 −0.002 2.084 −0.012 −0.010 −0.012 2.811 2.224 2.776 −0.003 2.173 −0.012 −0.012 −0.013 −0.012 2.271 2.713 2.834 2.753 −0.004 3.172 Notes. This table reports the results of the bivariate volume–volatility VAR. The variables are standardized with respect to duration and are in log-form to ensure that they remain positive. Estimation is based on the tick-by-tick data for the December 2008 expiry EUA futures contract traded on the ECX. The sample is 28th September 2005 to 15th December 2008. Only coefficients statistically significant at a significance level of 5% are reported along with absolute t-statistics. expected duration will fall and expected volumes will rise, making trade more likely.15 As a result of this rationale, the informed traders are associated with negative contemporaneous duration–volume and 15 Similarly, in Kyle (1985), insider (informed) traders act strategically. They rationally anticipate the impact of orders on prices, by conditioning on the behavior of other market participants (namely, the market maker and uninformed traders). As new private information arrives, informed traders review the existing price quotes of a specialist market maker and select the amount of shares which maximizes their informational advantage to trade. D. Bredin et al. / International Review of Financial Analysis 34 (2014) 222–234 227 Table 6 Trivariate duration–volume–volatility VAR (December 2008 contract). Duration equation Coefficient Current duration Lagged duration Current volume Lagged volume Current volatility Lagged volatility γ0 γ1 γ2 γ3 γ4 γ5 γ6 γ7 γ8 γ9 γ10 γ11 γ12 γ13 γ14 γ15 γ16 ρ0 ρ1 ρ2 ρ3 ρ4 ρ5 ρ6 ρ7 ρ8 ρ9 ρ10 ρ11 ρ12 ρ13 ρ14 ρ15 ρ16 δ0 δ1 δ2 δ3 δ4 δ5 δ6 δ7 δ8 δ9 δ10 δ11 δ12 δ13 δ14 δ15 δ16 Volume equation t-Stats 0.197 0.106 0.061 0.043 0.032 0.024 0.015 0.021 0.018 0.013 0.014 0.007 0.014 0.011 0.014 0.017 59.191 30.918 17.712 12.312 9.215 6.728 4.236 5.994 5.185 3.894 4.000 2.112 4.037 3.095 4.133 5.043 0.011 0.044 0.015 2.927 11.927 3.988 0.055 0.012 20.961 5.133 −0.006 2.321 −0.006 −0.007 2.213 −2.435 −0.007 2.754 Volatility equation Coefficient t-Stats Coefficient t-Stats 0.103 0.034 0.014 35.039 11.378 4.720 −0.008 2.686 0.399 −0.086 −0.038 −0.029 −0.016 −0.015 −0.018 99.939 20.668 9.127 6.941 3.723 3.552 4.161 −0.018 −0.009 −0.015 −0.013 4.185 2.076 3.530 3.165 −0.009 2.092 −0.114 0.018 0.016 0.013 25.889 4.097 3.568 3.000 0.205 0.067 0.044 0.026 0.033 0.018 0.017 0.021 0.027 0.018 0.012 0.015 0.016 0.014 0.013 0.019 64.123 20.441 13.391 7.916 10.019 5.542 5.062 6.301 8.105 5.493 3.772 4.489 4.771 4.283 3.961 5.816 −0.007 0.186 0.110 0.061 0.038 0.023 0.019 0.015 0.016 0.009 0.014 0.016 0.012 0.013 0.011 0.016 0.020 −0.060 0.023 2.212 58.191 33.872 18.681 11.514 7.093 5.804 4.661 4.770 2.922 4.273 4.842 3.686 3.962 3.423 4.994 6.304 25.889 9.637 Notes. This table reports the results of the trivariate duration–volume–volatility VAR with 16 lags. Estimation is based on the tick-by-tick data for the December 2008 expiry EUA futures contract traded on the ECX. The sample is 2nd January 2008 to 15th December 2008. Only coefficients statistically significant at a significance level of 5% are reported along with absolute t-statistics. duration–volatility relations and a positive contemporaneous volume– volatility relation. We adopt a causal ordering, consistent with the outlined rationale, that runs from duration, a variable with evident information content, to volume, a variable that also pertains to information content. In turn, either or both of these information laden variables may induce trades, which may move prices.16 Hence, the ordering of the variables in the bivariate system (and in the trivariate VAR) is theoretically informed.17 16 This is consistent with Manganelli (2005, p.382) who highlights ‘(i) the duration between trades can affect prices, (ii) duration should be correlated with volume, and (iii) both duration and volume should be correlated with transaction price variances.’ 17 Additionally, as a result of this rationale, it can be seen from Eq. (1), scaled volatility, zd,t and, volume, vd,t, are driven by two uncorrelated random shocks; an informed traded shock, ut and an uninformed traded shock, ϵt. 4. Data and empirical results 4.1. Data and contractual issues We adopt intra-day transaction data for Phase 1 and 2 EUA futures contracts with a December 2005–12 expiry.18 One futures contract corresponds to 1000 EUAs, and each EUA represents the right to emit 1000 tonnes of CO2 equivalent greenhouse gases. The settlement is three days after the last trading day. The data is taken from the continuous trading session, which runs from 7 am to 5 pm (UK local time). The 18 Contracts with a quarterly expiry exist (March, June, September and December), however the December expiry contracts are significantly more liquid and hence we focus on these instruments. For the remainder of the paper the contracts will be referred to by the year of expiration. 228 D. Bredin et al. / International Review of Financial Analysis 34 (2014) 222–234 Table 7 Trivariate duration–volume–volatility VAR (December 2008 contract — nearby contract). Duration equation Current duration Lagged duration Current volume Lagged volume Current volatility Lagged volatility γ0 γ1 γ2 γ3 γ4 γ5 γ6 γ7 γ8 γ9 γ10 γ11 γ12 γ13 γ14 γ15 ρ0 ρ1 ρ2 ρ3 ρ4 ρ5 ρ6 ρ7 ρ8 ρ9 ρ10 ρ11 ρ12 ρ13 ρ14 ρ15 ρ16 δ0 δ1 δ2 δ3 δ4 δ5 δ6 δ7 δ8 δ9 δ10 δ11 δ12 δ13 δ14 δ15 δ16 Volume equation Volatility equation Coefficient t-Stats 0.176 0.087 0.053 0.036 0.036 0.017 0.016 0.017 0.015 0.012 0.019 32.174 15.668 9.529 6.441 6.534 3.027 2.858 2.975 2.637 2.131 3.397 0.022 4.028 0.005 3.092 0.055 0.141 0.060 2.839 7.117 2.998 0.219 0.110 0.060 0.047 0.017 0.028 0.016 0.018 42.705 20.893 11.446 8.890 3.123 5.293 3.014 3.422 0.012 0.021 0.013 0.011 0.018 0.015 0.011 −1.091 0.300 2.326 4.020 2.444 2.066 3.351 2.812 2.111 25.615 6.809 0.091 2.085 0.105 2.417 2.001 12.365 Coefficient t-Stats Coefficient 0.022 0.004 0.005 15.494 2.457 3.462 0.011 −0.002 −0.001 −0.001 70.189 11.763 3.827 5.615 −0.001 2.495 −0.003 2.044 −0.001 2.390 −0.001 2.802 −0.016 0.005 −25.615 7.212 0.001 2.315 0.200 0.050 0.048 0.025 0.018 0.027 0.012 38.914 9.902 9.160 4.804 3.521 5.122 2.325 0.016 0.010 0.030 0.011 3.102 1.960 5.900 2.235 0.020 0.015 3.954 2.926 0.099 −0.448 2.308 t-Stats 2.791 Notes. This table reports the results of the tri-variate duration–volume–volatility VAR with 16 lags. Estimation is based on the tick-by-tick data for the December 2008 expiry EUA futures contract traded on the ECX. The sample is observed in November 2008. Only coefficients statistically significant at a significance level of 5% are reported along with absolute t-statistics. ECX has currently over 100 members trading EUAs.19 The minimum tick size is € 0.01 per tonne of CO2, while a range of order types are available including limit, market and stop orders.20 The annual fee for ECX members is € 2500, while the trading and clearing fee per contract is currently € 3.50. There are currently (introduced in December 2008) three market makers for the ECX (Five Rings Capital — formerly Jane Street Global Trading, Saxon Finance and RNK Capital). The market maker is required to provide bid and ask quotes for at least 85% of the trading time between 7 am and 5 pm (UK local time) and are required to respond to quote requests within 5 min. Our dataset includes comprehensive information relating to all transactions, maturities, trade prices, opening prices and trading volume for eight different futures instruments, 2005–12, traded on Phases 19 See www.ecx.eu for details. A dummy variable in respect to the intercept is adopted to take account of the change in the tick size, from € 0.05 to € 0.01, on the 27th March 2007 has been included. 20 1 and 2 of the EU ETS. To prepare the data for the analysis, we first sum the trading volumes when we detect a series of equal recorded prices for the same contract maturity which occur at the same time.21 The summation of trading volumes can be motivated by the practice in the management of the electronic book. This consists of utilizing the existing liquidity that is available in the sense that a trader can choose to split a large transaction into a series of smaller transactions and trading with different counter parties. We then compute the duration between trades, treating the overnight period as if it did not exist.22 21 We adjusted the data to remove diurnal patterns, which were found principally in the Phase II data, in line with the deterministic time filter of Tsay (2009). 22 As our futures contracts are quite thinly traded, following Manganelli (2005), we prefer to retain the overnight return. In respect to duration, we treat the overnight period as if it didn't exist. By way of contrast, eliminating absolute returns during the overnight period, e.g., by adopting a dummy variable, would have possibly caused the loss of important data for these quite illiquid futures contracts. D. Bredin et al. / International Review of Financial Analysis 34 (2014) 222–234 229 Table 8 Trivariate duration–volume–volatility VAR (December 2008 contract — signed volume). Duration equation Current duration Lagged duration Current volume Lagged volume Current volatility Lagged volatility γ0 γ1 γ2 γ3 γ4 γ5 γ6 γ7 γ8 γ9 γ10 γ11 γ12 γ13 γ14 γ15 ρ0 ρ1 ρ2 ρ3 ρ4 ρ5 ρ6 ρ7 ρ8 ρ9 ρ10 ρ11 ρ12 ρ13 ρ14 ρ15 ρ16 δ0 δ1 δ2 δ3 δ4 δ5 δ6 δ7 δ8 δ9 δ10 δ11 δ12 δ13 δ14 δ15 δ16 Volume equation Coefficient t-Stats 0.179 0.091 0.056 0.038 0.038 0.017 0.016 0.017 0.015 0.012 0.020 32.907 16.522 10.157 6.851 6.745 3.075 2.930 3.028 2.703 2.111 3.521 0.023 4.140 −0.078 5.195 1.935 12.054 Coefficient Volatility equation t-Stats 0.306 0.105 0.057 0.044 0.025 0.024 0.018 0.022 59.626 19.622 10.481 8.087 4.633 4.492 3.280 4.005 0.019 3.528 0.014 0.013 2.693 2.474 Coefficient t-Stats 0.011 −0.002 −0.001 −0.001 68.669 12.092 4.429 5.676 −0.001 2.345 −0.001 2.431 −0.001 2.779 0.001 1.930 0.197 0.052 0.048 0.026 0.019 0.019 0.013 38.413 9.902 9.131 4.933 3.664 3.664 2.434 0.018 0.010 0.029 0.010 3.500 1.969 5.756 2.059 0.020 0.014 3.877 2.911 Notes. This table reports the results of the trivariate duration–volume–volatility VAR with 16 lags. Estimation is based on the tick-by-tick data for the December 2008 expiry EUA futures contract traded on the ECX. The sample is 2nd January 2008 to 15th December 2008. Only coefficients statistically significant at a significance level of 5% are reported along with absolute t-statistics. 4.2. Empirical results In Table 1, we report the EUA futures contracts' descriptive statistics. The data examined for each contract varies considerably; with approximately 9, 19, 31 and 40 months available for the December 2005, 2006, 2007 and 2008 respectively. In respect to the 2009–12 contracts, we examine the calendar year of expiration exclusively, as these time periods relate to the majority of the volumes traded. The sample periods for our seven contracts are as follows: December 2005 (22nd April 2005 to 19th December 2005), December 2006 (16th June 2005 to 18th December 2006), December 2007 (3rd June 2005 to 17th December 2007) and December 2008 (28th September 2005 to 15th December 2008), December 2009 (2nd January 2009 to 14th December 2009), December 2010 (4th January 2010 to 10th December 2010), December 2011 (4th January 2011 to 19th December 2011) and December 2012 (3rd January 2012 to 17th December 2012). Indicative of the development of the market, particularly in the Kyoto Phase of the ETS, the number of transactions and the volume of trades grow approximately monotonically over time and in accordance the duration declines. Furthermore, the volatility in the Kyoto Phase is generally at a lower level than that in the Pilot Phase, with the exception of the heightened volatility present in the 2010 period. In Table 2, we examine the interaction over time between the variables in the VAR specifications and whether the theoretical justification for the ordering of the variables in the VAR can be rejected empirically using Granger causality tests. The distinction between the December 2005 contract and all other contracts (either Phase 1 or 2) is evident. We fail to reject a number of hypotheses for the December 2005 contract, for instance that volatility does not Granger cause duration and that either duration or volume does not Granger cause volatility, at 230 D. Bredin et al. / International Review of Financial Analysis 34 (2014) 222–234 the standard 5% level of significance.23 However, for the remaining contracts the results indicate strong inter-relationships between the three variables. Such strong evidence of inter-relationships and bidirectional causality between all variables does not provide evidence to counter our theoretical basis for ordering the VAR. Moreover, the bi-directional Granger causality points to evidence in favor of SIAH, and is inconsistent with the MDH, for the EU ETS futures market. Finally, these initial findings motivate further analyses, adopting VAR specifications, with regard to the interactions between trade duration, volume and volatility. Following Alberola et al. (2008) we take account of recognized structural breaks in the data, e.g., the compliance break in April 2006 and the break associated with the announcement of stricter validations of National Allocation Plans in October 2006, by including dummy variables to capture the potential shifts in the intercepts due to these breaks in the estimation of the 2006, 2007 and 2008 contracts. In addition, following Rotfuß (2009), we adopt a dummy variable in the intercepts to take account of a structural break March 27th, 2007 when the Intercontinental Exchange (ICE) reduced the minimum tick size for all EUA futures contracts from € 0.05 to € 0.01. It has been included in the model for the December 2007 and the December 2008 contract. We adopt an identical set of dummy variables across VAR specifications, where the sampled data contains the potential structural break. This latter dummy variable is statistically significant in all cases for the 2008 contract and there are mixed levels of significance for the 2007 contract. However, our results show that the impact of the breaks on volume and volatility is mainly confined to the 2006 contract. In each case the VAR model specification is estimated up to the optimal lag length identified by the AIC (akaike information criterion). The tables report significant coefficients only. 4.3. Phase I vector autoregression findings The estimates of the bivariate, duration standardized Pilot Phase, VAR specification represented in Eq. (1) are reported in Tables 3 and 4. In Table 3 the bivariate model for December 2005 is reported, while in Table 4 a summary of the Phase 1 results using the bivariate model is reported.24 The results (coefficient b0) reported in Table 4 indicate that there is a negative contemporaneous relationship between volume and volatility for all contracts. Moreover, the coefficient is highly significant in each case. A similar finding has been reported by Karpoff (1988) for the case of financial futures contracts, bonds and commercial paper. This relationship between volume and volatility is also contingent on the time duration between trades. The negative volume–volatility relationship is strengthened as the duration between trades increases since the contemporaneous interaction term between duration and volume (see coefficient c0 in Table 3) is negative and significant.25 This suggests that any empirical investigation of the intra-day volume–volatility relationship needs to actively account for the impact of time elapse between trades. The implication here, arising both from the negative contemporaneous relation between volume and volatility, as well as the strengthening of this relation, with the time elapse between trades, is that liquidity traders are dominating any informed trading behavior. 23 Notwithstanding, it is worthwhile highlighting that, despite a relatively small sample size as detailed in Table 1, the Dec 05 contract is associated with granger causality in three of the six bivariate relations studied [at the 5% level] and in four of the bivariate relations studied [at the 10% level]. Albeit this evidence isn't as convincing as the causality evidence found in respect to the 2006–11 contracts, it nevertheless is quite persuasive. 24 In the interests of conserving space, we have not reported the individual bivariate model results for all Phase 1 contracts. The individual contract results are qualitatively consistent and are available from the authors upon request. 25 The identified positive contemporaneous relation between duration and volatility, when controlling for volume, is indicative of the prevalence of liquidity traders in the market (Easley & O'Hara, 1992). The volatility is quite persistent, as evidenced by the positive significant coefficients on lagged volatility in Table 3. There is also consistent evidence for all contracts of positive autocorrelation in lagged volume. In addition, there is some evidence to indicate that lagged volume increases volatility, when accounting for duration effects. The estimated interactions over time between volatility and our lagged information variables, duration and volume, are consistent with the SIAH. Overall, the results are consistent across the Pilot Phase contracts. They indicate the relative prevalence of liquidity traders, considerable market development and the results also indicate the importance of the sequential information arrival hypothesis in the Pilot phase. 4.4. Phase II vector autoregression findings The estimates of the bivariate VAR specification represented by Eq. (1) for the Kyoto Phase are reported in Table 5 for the December 2008 contract. The estimated trivariate VAR specifications, represented in Eq. (2) are reported in respect to the December 2008 contract in Tables 6 to 8. The results reported in Table 5 are consistent with the SIAH and those results reported for Phase 1. There is a negative contemporaneous volume and volatility relationship which is highly significant. Table 9 A trivariate duration volume volatility VAR summary for December 2008–12 phase II contracts. Period 12 month VAR Nearby VAR Signed volume VAR Panel A: contemporaneous volatility–volume relation 2012 −0.01 −0.01 (t-value) (17.57) (4.62) 2011 −0.06 −0.02 (t-value) (36.13) (3.43) 2010 −0.10 −0.07 (t-value) (64.34) (11.54) 2009 −0.17 −0.08 (t-value) (72.53) (10.98) 2008 −0.11 −0.02 (t-value) (25.89) (25.62) −0.01 (4.56) −0.05 (13.37) 0.01 (7.38) 0.00 (3.24) – – Panel B: contemporaneous volatility–duration relation 2012 0.06 0.06 (t-value) (247.48) (75.49) 2011 0.13 0.14 (t-value) (237.04) (79.31) 2010 0.13 0.11 (t-value) (245.81) (58.14) 2009 0.14 0.14 (t-value) (217.00) (73.21) 2008 0.40 0.01 (t-value) (99.94) (70.19) 0.04 (41.87) 0.13 (235.61) 0.13 (242.11) 0.14 (210.9) 0.01 (68.67) Panel C: first order autocorrelation in volume 2012 0.23 0.22 (t-value) (116.63) (35.74) 2011 0.24 0.26 (t-value) (129.10) (47.30) 2010 0.18 0.22 (t-value) (92.11) (29.12) 2009 0.18 0.13 (t-value) (90.43) (5.38) 2008 0.19 0.22 (t-value) (58.19) (42.71) 0.26 (26.02) 0.43 (233.28) 0.16 (84.00) 0.43 (215.00) 0.31 (59.63) Notes. This table reports the summary results of the trivariate volume–volatility–duration VARs in respect to the Phase II contracts which expire December in each year 2008–11. In particular, it reports the contemporaneous volatility–volume coefficient, ρ0, the contemporaneous volatility–duration coefficient, γ0, in the volatility equation of these trivariate model specifications and the first-order autoregressive coefficient, δ0. The variables are standardized with respect to duration and are in log-form to ensure that they remain positive. Estimation is based on the tick-by-tick data for EUA futures contracts traded on the ECX. The December 2008 and 2009 contracts are sampled 2nd January each contract year to 15th December 2008 and 14th December 2009, respectively. The December 10 and 11 contracts are sampled from the 4th January of each contract year until 20th December 2010 and 19th December 2011, respectively. Finally, December 2012 is sampled 3rd January 2012 to 17th December 2012. Only coefficients statistically significant at a significance level of 5% are reported along with absolute t-statistics. D. Bredin et al. / International Review of Financial Analysis 34 (2014) 222–234 231 Fig. 1. Prices and trading volumes for Kyoto phase contracts. Notes. The figure presents EUA futures prices and cumulative trading volumes in respect to the December 2008, December 2009, December 2010, December 2011 and December 2012 expiration futures contracts. Further, this finding is strengthened as the duration between trades increases. The contemporaneous impact of the interaction term, duration by volume, on volatility is negative and significant across Kyoto Phase contracts. Finally, the bivariate VAR findings show lagged effects of information variables, volume and duration, on volatility.26 The main trivariate VAR specification findings for the December 2008–12 contracts are reported in Table 9.27,28 By treating duration as an endogenous variable, we are able to obtain a complementary understanding, relative to the bivariate specifications, of the interaction between time elapsed between trades, volume and volatility. Our results reported in Table 6 further highlight the level of consistency relative to the duration based bivariate VAR results. Volume continues to have a significantly negative impact on volatility contemporaneously. Moreover, the contemporaneous duration has a positive and statistically significant impact on volatility. As per the previous discussion, the positive relationship is also economically significant as it indicates that liquidity traders prevail in the EU ETS market rather than informed traders. In the trivariate model specifications, there is also a positive relationship between lagged volume and volatility which is consistent with the SIAH (see, Conrad et al., 2012). Again, consistent with the bivariate results as well as the SIAH, lagged duration has a negative and significant impact on current volatility. Trade arrival times, durations, convey private information in a similar manner to volume. A rational market maker will interpret high volumes in combination with short durations as evidence of trading by informed agents and will adjust quoted prices 26 To conserve space and given that the main emphasis is on the trivariate representation we do not report the bivariate results for the complete Kyoto phase. Full details of bivariate results for the Kyoto Phase contracts are available from the authors on request. 27 The Kyoto Phase estimates are for the year expiring samples. As can be seen from Fig. 1, the vast majority of the trading activity occurs in the contract year of expiration. Hence, all our Kyoto Phase analysis focuses on transaction data specific to the sample year for the particular contract, e.g. a January 2008 to December 2008 sample is adopted for the December 2008 contract. 28 We conduct similar analyses for each of the Pilot Phase contracts. The results are qualitatively similar to those obtained using the bivariate VARs in the Pilot Phase. Hence, to conserve space, we refrain from reporting the results here. The results are available from the authors on request. accordingly (Easley & O'Hara, 1992). As a result, within the SIAH framework, the lagged information variable, duration, is expected to be economically and statistically significant in the volatility equation of the trivariate VAR system. Given the significant evidence of lagged relationships, our results are at odds with the predictions of the MDH and are better characterized by the SIAH. In Fig. 2 we report the impulse response functions for the Kyoto Phase (Phase 2) 2008 December expiration contract for 25 lags (trades), including standard errors. The triangular decomposition (Cholesky) is adopted to achieve orthogonalization. The diagonal graphs indicate the effect of the shock on duration, volatility and volume on each of the respective terms. There is some persistence in observed variables, which is consistent with results in mature asset markets. For the offdiagonal entries, we find further evidence that duration does have an impact on volume and volatility, and vice versa, but generally over short time intervals.29 4.5. Robustness tests As a robustness test in respect to findings in the trivariate VAR model specifications, we estimate these models using data relevant for nearby months. Nearby contracts are defined in this market context as those trading in the month prior to their maturity, i.e. November for each contract. Our exclusion of observations on contract trading in the maturity month avoids biases associated with unusual market activity characteristic of times approaching maturity as traders unwind or rollover positions. Given the fact that a substantial proportion of the trading activity generally takes place in the contract specified on the nearby month, the analysis of data taken only from this contract data may be useful to further investigate our findings. This focus on the nearby month confines the data analysis to the most actively traded period 29 The results for the impulse response functions, in respect to the December 2005 contract, for 25 lags (trades), including standard errors, is reported in Fig. 3. Findings for the 2006 and 2007 contracts are available from the authors on request. The impulse response functions for Phases 1 and 2 are consistent, although the standard error bands are relatively larger for the earlier Phase 1 contracts given the relatively smaller number of observations. 232 D. Bredin et al. / International Review of Financial Analysis 34 (2014) 222–234 Fig. 2. Impulse response functions of the December 2008 contract. Notes. The figure presents impulse response functions, as well as the corresponding standard errors, for up to 25 lags for the December 2008 maturity futures contract. The sample period is 28th September 2005 to 15th December 2008. The graphs presented on the top left to bottom right diagonal of the figure trace out the time path of the effect of the shocks on duration, volatility and volume on each of the respective terms. The off-diagonal graphs trace out the time path of the effect of shocks across duration, volatility and volume. for each contract. The findings are reported in Table 7 in respect to the nearby months for the December 2008 contract, with the summarized findings presented in Table 8 for the complete set of Phase 2 contracts. In particular, volume has a contemporaneous significant negative impact on volatility and duration is contemporaneously positively and significantly associated with volatility. There is also some evidence that lagged volume increases volatility. Again, consistent with the bivariate results and the full blown trivariate results, lagged duration has a negative and significant impact on current volatility. As discussed above, taking these findings together they support the SIAH and they indicate that liquidity traders prevail in the current EU ETS market rather than informed traders. A second robustness test returns to the larger samples studied previously, however with signed volume substituted for volume. The results for the December 2008 contract are reported in Table 8, with the summary for all Phase 2 contracts reported in Table 9. The signed volume indicates whether the corresponding trade is a buyer initiated (+) or a seller initiated (−) trade while the conventional volume measurement which we adopt in the other specifications is always non-negative and does not convey this information. The contemporaneous negative relation between volume and volatility is no longer present. Notwithstanding, the contemporaneous duration has a statistically significant positive impact on volatility. The positive contemporaneous relation between duration and volatility is economically significant as it indicates that liquidity traders prevail in the current EU ETS market rather than informed traders. In addition, consistent with the bivariate results and the full blown trivariate VAR results, lagged duration has a negative and significant impact on current volatility. Finally, as is established by, inter alia, Lyons (1997) and Berger et al. (2005) for high frequency exchange rate data, pronounced positive signed autocorrelations of signed volume (i.e. order flow) results are evident at the tick-by-tick level in our study. This high level of autocorrelation may indicate the passing of a large trade among traders or it may reflect order splitting, whereby a large trade is executed sequentially. Given the significant evidence of lagged relationships, our results are once again at odds with the predictions of the MDH and are better characterized by the SIAH. The results reported in Tables 6–9 do not support MDH and they indicate the prevalence of liquidity traders rather than informed traders in the EU ETS market. Overall, the findings in both Pilot and Kyoto Phases of the ETS suggest, across contracts, sampled data and accounting for D. Bredin et al. / International Review of Financial Analysis 34 (2014) 222–234 233 Fig. 3. Impulse response functions of the December 2005 contract. Notes. The figure presents impulse response functions, as well as the corresponding standard errors, for up to 25 lags for the December 2005 maturity futures contract. The sample period is 22nd April 2005 to 19th December 2005. The graphs presented on the top left to bottom right diagonal of the figure trace out the time path of the effect of the shocks on duration, volatility and volume on each of the respective terms. The off-diagonal graphs trace out the time path of the effect of shocks across duration, volatility and volume. signed volume, the prevalence of liquidity traders and the importance of the SIAH. liquidity trades are dominating any informed trader behavior. Informed traders, as is often the case, may have a preference to trade on the over the counter (OTC) market. 4.6. Carbon finance market implications 5. Conclusions Our detailed high frequency examination of the EU ETS extends the large number of recent studies examining carbon finance market developments using low frequency data. Exchange traded market activity, and EUA futures in particular, have grown dramatically with the move from Pilot to the Kyoto Phase. Besides the spike in 2006, mean duration and volumes for the contracts examined have been reflective of a market that is developing year on year. When we take account of trade initiation using signed volume our results continue to indicate a market being driven by buyer initiated trades rather than seller initiated trades. Our conclusion is supported by findings in a recent study by Kalaitzoglou and Ibrahim (2013). The authors highlight that the EU ETS can be viewed as a buyer orientated market, where a greater proportion (in terms of volume) of trades are buyer rather seller initiated. Our empirical results for duration and its relation with both volume and volatility are consistent with this view. The implications are that This paper investigates the initial stages of the European Union (EU) emission trading scheme (ETS) introduced formally in January 2005 from a market microstructure perspective. The scheme instigated as part of the EU agreement to cut worldwide emissions of carbon dioxide (CO2) issues a restricted amount of emission allowances to firms on an annual basis and subsequently allows firms to trade the emission permits they hold. This study covers the first phase of trading from 2005 to 2007 and the period from 2008 to 2012 of the second phase, which coincides with the first compliance period of the Kyoto Protocol. The main motivations for making inferences is to better understand the information assimilation process, liquidity and the degree of market efficiency during the phases of the emission trading scheme. We examine the microstructure and trading behavior, in the ETS, in the context of these objectives. 234 D. Bredin et al. / International Review of Financial Analysis 34 (2014) 222–234 Our key result is that significant market developments have taken place, with reduced market frictions suggesting evidence of greater levels of efficiency along the lines discussed in Dacorogna, Gencay, Muller, Olsen, and Pictet (2001). This is reflected in the form of reduced durations between trades, higher volumes traded and lower volatilities. Secondly, bi-directional Granger causality is found in the trade variables, the empirical evidence is consistent with the sequential information arrival hypothesis (Copeland, 1976) rather than the predictions of the mixture of distribution hypothesis (Clark, 1973; Tauchen & Pitts, 1983). Ours is the only study to take account of duration in relation to the sequential information arrival versus the mixture of distribution hypothesis for the EU ETS. In addition, our evidence in favor of the sequential information arrival hypothesis is consistent with recent results on the EU ETS reported by Conrad et al. (2012) and Chevallier and Sevi (2011). Finally, our results indicate a negative contemporaneous relationship, accounting for duration effects, between volume and volatility. The implication is that liquidity traders dominate any role played by informed traders (Easley & O'Hara, 1992). Notwithstanding the significant developments, this result is not surprising for a market that is dominated by hedgers, see Kalaitzoglou and Ibrahim (2013). A common finding from the VAR models indicates that there is a negative contemporaneous relationship between volume and volatility for all contracts. The implication here is that liquidity traders are dominating any informed trading behavior. This result is consistent with the market view that informed traders, as is often the case, may have a preference to trade on the OTC market. If informed traders are active in the OTC market, is this due to greater levels of depth, breadth and overall market liquidity? We leave this question to be examined by future research. References Alberola, E., Chevallier, J., & Cheze, B. (2008). Price drivers and structural breaks in European carbon prices 2005–2007. Energy Policy, 36(2), 787–797. Andersen, T. (1996). Return volatility and trading volume: An information flow interpretation of stochastic volatility. Journal of Finance, 50, 169–204. Bagehot, W. (1974). The only game in town. Financial Analysts Journal, 27, 12–14 (22). Benz, E., & Hengelbrock, J. (2009, May 5). Liquidity and price discovery in the European CO2 futures market: An intraday analysis. Carbon Markets Workshop, LSE. Berger, D. W., Chaboud, A. P., Chernenko, S. V., Howorka, E., Iyer, R. S. K., Liu, D., et al. (2005). Order flow and exchange rate dynamics in electronic brokerage system data. IFDP 830. Board of Governors of the Federal Reserve System. Blume, L., Easley, D., & O'Hara, M. (1994). Market statistics and technical analysis: The role of volume. Journal of Finance, 49(1), 153–181. Bredin, D., & Muckley, C. (2011). An emerging equilibrium in the EU Emissions Trading Scheme. Energy Economics, 33(2), 353–362. Chevallier, J. (2011). Macroeconomics, finance, commodities: Interactions with carbon markets in a data-rich model. Economic Modelling, 28(1–2), 557–567. Chevallier, J., & Sevi, B. (2011). On the realized volatility of the ECX CO2 emissions 2008 futures contract: Distribution, dynamics and forecasting. Annals of Finance, 7, 1–29. Clark, P. K. (1973). A subordinated stochastic process model with finite variance for speculative prices. Econometrica, 41(1), 135–155. Conrad, C., Rittler, D., & Rotfuß, W. (2012). Modelling and explaining the dynamics of European Union allowance prices at high frequency. Energy Economics, 316–326. Convery, F. J., & Redmond, L. (2007). Market and price developments in the European Union Emissions Trading Scheme. Review of Environmental Economics and Policy, 1(1), 88–111. Copeland, T. E. (1976). A model of asset trading under the assumption of sequential information arrival. Journal of Finance, 31, 1149–1168. Dacorogna, M. M., Gencay, R., Muller, U., Olsen, R. B., & Pictet, O. V. (2001). An introduction to high-frequency finance. USA: Academic Press. Diamond, D., & Verrecchia, R. E. (1987). Constraints on short-selling and asset price adjustment to private information. Journal of Financial Economics, 18(2), 277–311. Dufour, A., & Engle, R. F. (2000). Time and the price impact of a trade. Journal of Finance, 55(6), 2467–2498. Easley, D., & O'Hara, M. (1992). Time and the process of security price adjustment. Journal of Finance, 47, 557–605. Ellerman, A.D., & Buchner, B. (2006). Over-allocation or abatement? A preliminary analysis of the EU ETS based on the 2005 emissions data. Nota di Lavoro 139.2006. FEEM (also listed as WP-2006-016 at MIT CEEPR). Engle, R. F. (2000). The econometrics of ultra-high frequency data. Econometrica, 68, 1–22. European Environmental Agency (2012, October). Greenhouse gas emission trends and projections in Europe 2012. Harris, L. (2003). Trading & exchanges: Market microstructure for practitioners. Oxford University Press. Kalaitzoglou, I., & Ibrahim, B.M. (2013). Does order flow in the European carbon allowances market reveal information? Journal of Financial Markets, 16, 604–635. Karpoff, J. M. (1988). Costly short sales and the correlation of returns with volume. Journal of Financial Research, 11, 173–188. Kyle, A. S. (1985). Continuous auctions and insider trading. Econometrica, 53, 1315–1335. Lyons, R. (1997). A simultaneous trade model of the foreign exchange hot potato. Journal of International Economics, 42, 275–298. Manganelli, S. (2005). Duration, volume, and the price impact of trades. Journal of Financial Markets, 8, 377–399. Mansanet-Bataller, M., Pardo, A., & Valor, E. (2007). CO2 Prices, Energy and Weather. The Energy Journal, 28, 73–92. Mansanet-Bataller, M., & Pardo, A. (2008). CO2 prices and portfolio management. University of Valencia working paper (http://ssrn.com/abstract=1156975). Mizrach, B. (2010). Integration of the global emissions trading markets. Rutgers University working paper (http://ssrn.com/abstract=1621340). Mizrach, B., & Otsubo, Y. (2014). The market microstructure of the European climate exchange. Journal of Banking and Finance, 39, 107–116. O'Hara, M. (1987). Trading mechanisms and stock returns: An empirical investigation: Discussion. Journal of Finance, 42, 554–555. Paolella, M. S., & Taschini, L. (2008). An econometric analysis of emission trading allowances. Journal of Banking and Finance, 32, 2022–2032. Redmond, L., & Convery, F. J. (2006). Determining the price of carbon in the EU ETS. Planning and environmental policy research series (PEP) working paper 06/09. School of Geography, Planning and Environmental Policy, University College Dublin. Rittler, D. (2009). Price discovery, causality and volatility spillovers in European Union allowances phase II: A high frequency analysis. Discussion paper no. 492. Department of Economics, University of Heidelberg. Rotfuß, W. (2009). Intraday price formation and volatility in the European Emissions Trading Scheme: A first analysis. Working paper. Mannheim: ZEW. Tauchen, G. E., & Pitts, M. (1983). The price variability–volume relationship on speculative markets. Econometrica, 51, 485–505. Tsay, R. (2009). Conditional duration models. In T. C. Mills, & K. Patterson (Eds.), Handbook in econometrics, vol. 2. (pp. 1004–1024). Palgrave (Chapter 21). Xu, X., Chen, P., & Wu, C. C. (2006). Time and dynamic volume–volatility relation. Journal of Banking and Finance, 30, 1535–1558.