Statistical Arbitrage and Information Flow in an Electricity Balancing Market.

• Author: Bunn, Derek W.

1. INTRODUCTION

   In progressing towards more efficient competitive electricity markets, the liberalizing intent has generally been to replace central control with price signals and markets wherever possible. This is becoming the norm in forward, day ahead, and intraday trading, but in the provision of real-time balancing, progress in this direction has been more cautious. By "real-time" activities we refer to the actual production and consumption of electricity and its instantaneous matching on the network (i.e. "balancing") by the system operator. Prior to this physical delivery to meet demand, all other electricity trading is in forward commitments. Balancing arrangements tend to be idiosyncratic in details, according to system operations and jurisdictions around the world, but a common element is that at some point in time ahead of real-time delivery, forward trading between market participants has to stop and participants must nominate to the system operator their physical intentions for production and consumption over the subsequent delivery period. In many markets this represents the point ("gate closure") at which voluntary bilateral forward trading between producers and retailers stops and the system operator becomes counter-party to any further real-time trades; in other markets where mandatory engagement in optimized dispatching procedures is required (e.g. markets setting locational marginal prices) this is the point at which final dispatch instructions are received. In either case, thereafter, there will usually be a real-time "participant imbalance", defined as the difference between a prior nomination (to produce or consume) and the subsequent metered volume, and the aggregate of all of these, the "system imbalance" must be managed through the real-time balancing directives of the system operator. All participant imbalances must be settled financially ex post via administered "imbalance prices".

   This terminology for "imbalances" is becoming quite widespread in the industry. The mechanism for setting the imbalance prices is evidently not a market in the sense that the prices emerge endogenously from producers and consumers agreeing to trade between each other, but they are administered following well-defined procedures by the system operator on the imbalance volumes of the participants, and may, depending upon the arrangements, constitute costs or revenues to the participants. Conventional wisdom used to be that, for system control purposes, market participants should be obliged to keep to these real-time nominations, either through central control or motivated to do so by the application of penalties on their imbalance volumes. However, it is an open question if further liberalization, involving a relaxation of this obligation in order to permit or even encourage a degree of participant imbalance would be beneficial, and if so, how might market participants manage their operations accordingly.

   The economic efficiency argument for this relaxation is appealing. If the settlement price faced by the participants for their imbalances reflects the actual-real time cost of the balancing actions taken by the system operator, market participants that correctly anticipate the direction of system imbalance for the whole system can profit by being out of balance in the opposite direction. Thus, if the system as a whole is short (i.e. actual load is higher than expected from the aggregate participant nominations) and the system operator is therefore buying balancing energy, the imbalance settlement price of the system will be higher than the forward market prices, and so a generator that "spills" extra power in the delivery period will receive this higher imbalance settlement price for its imbalance. The situation is similar for active consumers and analogously beneficial in reverse when the system imbalance is long. By attempting to profit from "opportunistic imbalancing," active participants would seek to correctly anticipate the direction of system imbalance and would effectively be undertaking a form of statistical arbitrage (i.e. trading on the basis of statistical expectations). If successful (and permitted), they would thereby gain balancing revenues for themselves and reduce the balancing needs of the system operator.

   Such speculation is risky and raises the question of what aspects of market design render the risk-return payofs for participants taking these actions to be profitable. Beyond that question are the obvious system operations concerns that if there were to be excessively synchronized multi-agent responses to real-time price signals, lagged to some extent by information flows, the stability of the network may be put at risk. In this paper, therefore, we analyse the incentives for both asset-backed physical traders and non-physical ("virtual") traders to engage in this statistical arbitrage through opportunistic imbalancing, how this might become more profitable as short-term wind and solar forecast errors increase the level of system imbalances faced by the system operator, and whether these activities can reduce the overall balancing costs to the system operator whilst limiting any potential detrimental effects on system instability. Note for clarity that this "opportunistic imbalancing" is distinctly not a market power strategy by a large player influencing quantities and prices in the spot market; rather it is a more benign price-taking response by any player, large or small, to an expectation that the system operator for the market as a whole will be buying or selling in real-time. This is not envisaged as a strategic game between a market participant and the system operator.

   The first pre-condition to attract this statistical arbitrage is evidently a single price settlement process, ie for each real-time delivery (imbalance settlement) period, the system operator defines one imbalance price that will be applied for settlements in both directions: if a participant is spilling it will receive this price for its imbalance volume whereas if it is in deficit it will pay this price for its imbalance volume. In markets where real-time balancing is undertaken through repeatedly optimized dispatching adjustment algorithms, eg every 5mins as in California, New York and various other ISOs, these final-run, real time prices are naturally used as the basis for settlement (Wang et al., 2015). In other markets, the system operator may be calling upon reserves for up or down regulation minute by minute, and/or accepting bids and ofers from generators and consumers continuously during each delivery period, eg in Britain and Germany.

   In the continuous case, the system operator may be taking balancing actions in opposite directions many times during each delivery periods. In these circumstances, the duration of imbalance settlement periods may be hourly, half hourly or 15 mins and so identifying a single imbalance price for settlement is usually quite detailed and specific. Usually some procedure is followed to identify the net imbalance volume over the imbalance settlement period and estimate the corresponding marginal cost of the system operator's balancing actions to determine the imbalance price (e.g.. Elexon, 2016). Figure 1 shows the relationship of the imbalance prices in GB in July 2019 (called "system prices" by the System Operator) to the net imbalance volumes in each half-hourly delivery period. Depending upon whether the system is short (positive imbalance) or long (negative imbalance) the prices show distinct distributions. Fundamentally for a short market, the price will be above marginal cost (since the system operator is buying marginal power), whilst for a long market, the price will be below marginal cost (since the system operator is selling power to reduce marginal production).

   In contrast, several markets use dual settlement prices derived from this continuous process to represent the different average, or marginal, costs of buying and selling actions by the system operator during each delivery period (e.g. France, Spain, Italy). Dual settlement prices are generally applied in order to deter the opportunistic imbalancing, as indicated above. In this case, to the extent that the system operator's buying/selling prices will be above/below the market price before delivery, it would always be better for a participant to buy/sell in the forward market pre-delivery, rather than seeking to be imbalanced during the delivery period.

   To motivate our analysis of the benefits or otherwise of statistical arbitrageurs operating in the balancing and settlements process, we record a "natual experiment" in the progressively liberalized evolution of balancing arrangements in the British wholesale market. When the British market was first liberalized in the 1990s, central control was retained with a mandatory day-ahead competitive auction providing an algorithmic unit commitment and disptach, followed by socialized balancing costs recovered through an "uplift" in the wholesale purchase costs for the retailers. When the trading was further liberalized to voluntary bilateral transactions in 2001, dual settlement prices were initially implemented to incentivize self-balancing by market participants.

   Shortly afterwards, a modification was introduced to avoid penalising those imbalances that were beneficial to the system (i.e. participant imbalances counter to the direction of system imbalance) by having one of the dual settlement prices revert to the pre-delivery, forward market price if it were in the counter-direction to the system. Whilst avoiding a penalty, this still did not incentivise opportunistic imbalancing. Then in 2015, the dual settlement system was changed to a single price, mainly to provide a clearer signal for the provision of flexible reserve capacity and innovative services. In proposing this, the regulatory body noted that the previous dual pricing 'drives inefficiency in...