آیا جریان سفارش در بازار آتی کربن اروپا اطلاعات را آشکار می کند؟

This paper identifies the classes of agents at play in the European Carbon Futures Market and analyzes their trading behaviour during the market's early development period. A number of hypotheses related to microstructure are tested using enhanced ACD models. Evidence is presented that the market is characterized by three different groups of traders: informed, fundamental, and uninformed. OTC trades are distinct to regular trades and are used strategically by the informed. Fundamental traders react faster in Phase II and the informed counteract by increasing their trade size and speed. The results indicate enhanced market transparency and increased market maturity.

Several studies have analysed the microstructure of the European Carbon Futures Market that developed with the implementation of the European Union Emission Trading Scheme (EU ETS). Benz and Hengelbrock (2008) is the first study on price leadership and discovery; Paolella and Taschini (2008), Benz and Trück (2009), and Conrad, Rittler, and Rotfuss (2012) investigate the relation between price formation and volatility; Mizrach and Otsubo (2011) analyse price discovery across spot and futures markets and the predictive content of order imbalances; Frino, Kruk, and Lepone (2010) examine the relation between liquidity, trading costs and volatility, and Ibikunle, Gregoriou, and Pandit (2011) study liquidity variations around the new rules of the Kyoto commitment period (2008–2012). Other contributions are largely along these lines. Zhang and Wei (2010) provide a survey of the literature on the market's operating mechanism and economic effect. In particular, apart from Bredin, Hyde, and Muckley (2011), who focus on the volume–volatility relationship, little has been done on the identification of trading behaviour, especially through non-price related order flow variations and, specifically, through the modelling of the time between trades (duration). The typical market microstructure model assumes two classes of agents: informed traders who possess private information about future values and uninformed traders who are primarily liquidity motivated (c.f., Madhavan, 2000). Admati and Pfleiderer (1988) further dissect the uninformed class into discretionary traders who time their trades and non-discretionary traders who arrive randomly. They show that the strategic play between the informed and the discretionary traders generates bouts of trading reflected in clusters of high trading activity. An appropriate time series analysis of trade characteristics, such as trade sign, size, frequency, and timing, should then reveal the presence, type, and magnitude of impact of the various agents at play. Diamond and Verrecchia (1987) and Easley and O’Hara (1992), for example, show that the time between trades (duration) is related to information. Specifically, longer durations can be associated with either a specific type of news or the absence of news. Dufour and Engle (2000) explicitly incorporate this time dimension into the pricing model of Hasbrouck (1991) using the Autocorrelated Duration (ACD) model of Engle and Russell (1998). They provide evidence that trade duration is informative and high trading frequency reflected in short duration is associated with a high price impact, faster price adjustment to new information, and stronger autocorrelation of trades. Beside the important implications this has on prices, it is evidence that non-price characteristics of trades carry information that may enable the identification of the various agent classes at play during trading. As a transaction on its own does not reveal the class of agent that initiated it, its characteristics relative to those of adjacent others might. Methodologically, if trading behaviour is revealed in trade characteristics, such as duration, then it is reasonable to expect these characteristics to cluster in regimes each having a different distribution. Accordingly, that specific trade characteristic would be described by a mixture of distributions. Hujer and Vuletic (2007) suggest the introduction to the ACD framework of an underlying ‘unobservable’ mixing variable that “culls the presence of unobservable information regimes and the mixing parameters pose as fractions of different information regimes…”. Through this theoretical hinge they are able to link the hazard rate of duration (instantaneous transaction rate) to classes of agents. This paper identifies the classes of agents at play in the European Carbon Futures Market and analyzes their trading behaviour during the market's early development period. A number of hypotheses relating to microstructure are tested. These are posed in the form of questions. Does clustering of non-price trade characteristics correlate with the trading behaviour of agents? If so, then how many classes of strategic agents are at play in the Carbon Market? Would trading intensity, defined as volume weighted duration, which can act as an observable, as opposed to Hujer and Vuletic's (2007) latent, ‘mixing’ variable, carry sufficient information for the identification of agents on a trade-by-trade basis? Are agents' actions, as displayed by non-price trade characteristics, sufficiently distinct to be revealed in different regimes of duration with different distributions? If so, then do these agents' learning speeds differ, especially as learning speeds can be captured by a smooth transition mechanism between different regimes of duration. Further, if the class of informed traders can be identified, do they tend to act at once in a large volume as in Glosten and Milgrom (1985), or stealthily in a segmented fashion as in Kyle (1985)? Furthermore, what is the role of OTC trades in information revelation? Are OTC trades dominated by specific classes of agents? The expectation is that larger informed trades may have preference in being initiated over-the-counter due to economies of scale, negotiating flexibility, and other advantages. Finally, how do the preceding queried characteristics of trading behaviour vary over the trading day and across the two main trading environments of the European Climate Exchange (ECX) and Nord Pool (NP)? For example, do certain classes of agents dominate trading during certain times of the trading day, or in ECX over NP? These questions motivate the analysis. This paper uses enhancements to Hujer and Vuletic's (2007) methodological advancement to answer these questions. Transaction duration is modelled with the smooth transition mixture of Weibull ACD models. The particular enhancements used have certain advantages. First, they allow for testing whether volume and duration can, when combined, act as an ‘observable’ variable sufficient to identify classes of agents. Second, the smooth transition formulation allows for capturing different learning speeds across agents.2 Third, the incorporation of an OTC dummy distinguishes the effects of OTC trades in the duration process, allowing for a more accurate description of trading behaviour and means by which the role of these trades can be investigated. The results shed light on the hypotheses posed above. The contribution of this paper is three-fold. To our knowledge, it is the first detailed analysis of trading behaviour directly aimed at identifying classes of agents at play in the European Carbon Futures Market. It is also the first that models duration in this market through a regime threshold ACD specification, and one that is motivated by three, rather than two, distinct behavioural patterns.3 Strong evidence is presented of three classes of agents at play with distinct trading patterns discussed in the microstructure literature: informed trades characterized by high volume and low duration, uninformed or non-discretionary liquidity trades that arrive randomly, and fundamental or discretionary liquidity trades that have lagged behaviour. Most previous literature has analysed only two regimes and in other markets. Evidence is also presented that the interplay between the informed and the fundamentals, first alluded to by Admati and Pfleiderer (1988), is present in the carbon market. Methodologically, the adopted model specification enhances that of Hujer and Vuletic (2007) in three aspects: First, the shape of the error distribution is data driven rather than being predetermined, which allows the data to freely indicate trading behaviour. Second, the threshold variable (trading intensity) is observable rather than latent, which is directly related to trade characteristics and, hence, is easier for traders to identify and measure. Finally, the introduction of a smooth transition mechanism captures hybrid trades and behaviour indeterminate between regimes. The model also enhances that of De Luca and Zuccolotto (2006) by recognizing three regimes rather than a restrictive two and that each regime can adopt a distinctly different distribution for the errors rather than merely different shapes of the same distribution, which has implications on the identification of distinct trading behaviour, especially that of discretionary liquidity traders. The second contribution is the analysis of the learning speeds of agent classes. The adoption of a smooth, rather than a discrete, transition mechanism reveals new evidence of changing learning speeds of participant groups across phases of the carbon market. The fundamentals learn faster in the Kyoto commitment period of Phase II (2008–2012) and this forces the informed to submit larger trades. The third contribution stems from recognizing the distinct role of OTC trades. Their distinct characteristics are analysed in both the ECX and the NP markets, over the early years of the market development and throughout the trading day. Incorporating their characteristics into the modelling of duration reveals that these trades are heavily used by informed traders, contribute significantly to information revelation, and have a large impact on liquidity with implications on market depth. The remainder of this paper is organized as follows. Section 2 provides a background to the Carbon Futures Market, Section 3 sets the microstructure framework that underlies the analysis, Section 4 presents the adopted methodology, Section 5 discusses the results, and Section 6 summarizes and concludes