Speculators and Price Inertia In A Day-Ahead Electricity Market: An Irish Case Study

1. Introduction

Electricity markets have become a prominent area of academic research in recent decades driven by transformative developments such as market deregulation, the rapid expansion of renewable generation and the growing availability of short-term trading opportunities. These changes have spurred research across diverse areas including market power and competition [1,2], price volatility [3], electricity price forecasting [4,5,6,7], portfolio and asset optimisation [8], and market dynamics [9]. In many jurisdictions the Day-Ahead (DA) market is one of the main routes to market through which electricity participants hedge or rebalance their positions ahead of physical delivery. DA market prices serve as reference prices for a range of financial derivative contracts and also influence retail electricity pricing. Given its central role, further insights into the functioning of DA markets remain important. This paper contributes to that effort by examining two distinct but interrelated topics within a DA market setting: the activity of financial traders and price inertia, the latter defined as the extent to which market prices resist small shifts in demand or supply. A deeper understanding of these aspects is relevant for regulators evaluating market design and liquidity, as well as for participants active in short-term electricity markets.

As noted by [10], the role of financial traders in commodity and energy markets has been examined in a range of studies. In the context of short-term electricity markets, we refer to financial traders as speculators, defined as participants that neither generate nor consume electricity but instead earn profits or incur losses based on the spread between their buy and sell prices either within the same market or across different market timeframes. In several U.S. electricity markets speculators can engage in Virtual Bidding, a mechanism that enables them to profit from the spread between DA and Real-Time (RT) prices1. This topic has been examined in studies such as Parsons et al. [2015], which presents cases where Virtual Bidding narrows the DA–RT spread but offers limited system benefits. In contrast, [14] find that Virtual Bidding can enhance liquidity and transparency, reporting a reduction in both the spread and its volatility in California’s wholesale market. In [15] it is seen that lower transaction costs in MISO were accompanied by increased speculator participation, altered generator behaviour, and improved consumer welfare. Meanwhile, Birge et al. [2018] observe that speculators may reduce forward premiums (i.e the DA-RT spread), although their impact is constrained by capital and regulatory limits2. While this U.S. focused literature offers valuable insights into price spreads and market efficiency, our study addresses a different and underexplored question: the presence and commercial behaviour of speculators in a European DA market. In contrast to the definitional ambiguity found in some of the broader commodity and energy market literature alluded to by [10], our analysis offers a clear and transparent framework for identifying speculators and quantifying their presence in a short-term European electricity market setting.

While prior research has examined short-term electricity market price volatility and the occurrence of price spikes, relatively little attention has been given to price inertia in these settings. Studies addressing volatility include, among others, [17,18,19,20]. In particular, [20] analyse demand and supply shocks in the Nordic market and show that price jumps are more likely when the system is operating near capacity constraints. In terms of price inertia in DA markets, studies such as [21] and [22] touch on the concept indirectly, as their methodologies examine the effect of larger supply shifts on prices. In a DA electricity price forecasting context, [23] present a single trading period example illustrating how steep supply or demand slopes can lead to high price sensitivity to external shocks, a similar intuition underlies our approach to measuring price inertia. One rare example of work that explicitly considers market stability is [24] which finds that increased integration and competition in the Nordic market led to more stable and less volatile prices. While conceptually related to our analysis, their approach differs methodologically and uses daily average prices, whereas we focus on hourly price data. In this paper, we propose an intuitive and easily understood method for quantifying price inertia in DA markets. Our approach draws conceptual inspiration from liquidity measures and depth analysis techniques commonly encountered in continuous trading markets, tools that are familiar to many practitioners, but is developed specifically for an auction-based setting.

1.1. Research Questions

This paper addresses two distinct but related research questions:

• What is the extent of speculator participation in the market, and how has it evolved over time?
• Can we define a clear and intuitive approach to measuring price inertia, and how has it changed over time?

To answer the first question, we use granular participant-level order and trade data to quantify the share of overall market activity attributable to speculators and to measure their influence on price setting, specifically how often their bids or offers equal the market-clearing price (i.e., marginality3). We also examine how speculator behaviour changed following a structural market change. To answer the second question, we introduce a sensitivity-based method for quantifying DA price inertia using publicly available hourly demand and supply curves, capturing how market prices respond to small shifts in supply or demand.

Our findings show that speculators are frequently marginal in the DA auction despite representing a small share of total market activity. We also observe a shift in price inertia following the structural market change.

This research contributes to the literature by providing additional insights into the functioning of DA electricity auctions. While the role of speculators in influencing price spreads has been studied extensively in U.S. short-term electricity markets, we are not aware of any empirical studies focusing specifically on speculators in European auction-based DA markets. This paper presents evidence that speculators are a significant presence in the market, particularly in terms of marginal pricing. Our second contribution is the development of an intuitive and transparent methodology for quantifying price inertia in DA markets. This is especially relevant in the context of ongoing market design and reform discussions, such as Great Britain’s Review of Electricity Market Arrangements [29,30]. Although price volatility has been well studied, its underlying drivers, such as low price inertia, remain comparatively underexplored ([24] presents a similar motivation in their work on market stability). Price inertia, as an additional market metric, can inform the evaluation of market design proposals by offering insight into how sensitive prices are to small shifts in demand or supply, an important consideration when evaluating past, present and potentially future price dynamics. The appeal of our approach lies in its conceptual clarity and ease of implementation. It offers a new analytical lens, similar in spirit to liquidity or order book depth analysis in continuous markets, but developed specifically for an auction-based setting. In addition to its practical applicability, our analysis provides concrete evidence on how DA price inertia evolved before and after a structural market change. While the methodology we propose is purposefully straightforward and transparent, it is informed by the complexity of the market environment, which makes an observational, data-driven approach both appropriate and necessary. These complexities, along with the key simplifying assumptions underlying our analysis, are discussed in subsequent sections.

1.2. Paper Structure

The structure of the remainder of the paper is as follows: in Section 2 we present both an overview of the Irish short-term electricity market and related pricing algorithms, we also provide a hypothetical example of a speculator in this setting. In Section 3 we detail the data sources and methodology. Section 4 presents the empirical analyses whilst in Section 5 we discuss the main findings. In Section 6 we conclude. Supplementary details and materials are available in the accompanying appendices.

2. Market Structure

2.1. Market

The Integrated Single Electricity Market (I-SEM), is the electricity market in Ireland. Although it is a small market (net electricity consumption in Ireland in 2021 was 30TWh, the equivalent figure for Great Britain was 305TWh [31]) it typifies spot European wholesale electricity market structures. The Irish electricity market is an example of what is called a multi-settlement electricity market, that is where the bidding and dispatch of electricity for a specific trading period is managed in successive runs. Some of the routes to market, that is how to buy and sell electrical energy, include

• Day-Ahead (DA) market: at 11am on day D participants submit orders to buy or sell electricity for hourly delivery periods in the [11pm D, 10pm D+1] interval. The market coupling algorithm, EUPHEMIA (Section 2.2), takes these inputs, in conjunction with interconnector transmission capacities (and other factors), and determines hourly prices and the direction of energy flow on the interconnectors. If the network is congested then zonal prices will diverge.
• Intraday (ID) market: after the DA auction has cleared additional auctions are held; these auctions, IDA1, IDA2 and IDA3 are similar in nature to the DA with the main differences being that they are held closer to the delivery time and the delivery periods are 30 minute intervals4.
• Intraday Continuous (IC) market: this is an important market in other European jurisdictions, but in an Irish electricity market context it comprises less than half a percent of traded energy volumes over the study timeframe and hence it is ignored here.
• Balancing (B) market: one hour prior to delivery ex-ante5 trading opportunities cease and the Transmission System Operator (TSO), takes over. The TSO compares forecast demand versus forecast generation, including what volumes have been traded in the ex-ante markets, and dispatches on/off units to ensure that supply meets demand. The prices that result from these dispatch decisions are called settlement imbalance prices (or Balancing market prices).

Securing a physical position in the Irish electricity market requires participation in one of the above markets. The market structure therefore provides a non-trivial level of (DA) liquidity. Again, for clarity purposes we note that speculators that are active in the Irish electricity market do not consume or generate electricity.

Structural Market Change

For the study timeframe the electricity markets in Ireland and Great Britain were connected via two interconnectors. Prior to Great Britain leaving the EU (i.e. Brexit), the these markets were coupled in DA and IDA1 auctions; post 1st January 2021 the coupling now takes place at the IDA1 and IDA2 auctions [32,33].

2.2. Pricing Algorithm

This section provides a brief description of the DA market pricing algorithm. It will be seen that the algorithm is more complex than a simple intersection of supply and demand curves and a complete description (including implementation details) is not publicly available. This combination of complexity and relatively limited transparency is one reason we adopt a reasonably straightforward observational approach when analysing speculator behaviour (Section 3.2). We also note that our price inertia analysis (Section 3.3) relies on a key simplifying assumption that abstracts from the full market-clearing process.

EUPHEMIA, the Pan-European Hybrid Electricity Market Integration Algorithm, is the algorithm used to calculate hourly DA market prices in a number of European markets. The EUPHEMIA Public Description [34] explains

• "The algorithm can handle a large variety of order types at the same time"; these include Aggregated Hourly Orders, Complex Orders (including Minimum Income Condition, MIC, and/or Load Gradient constraints), Scalable Complex Orders and Block Orders.
• The algorithm solves a Welfare Maximization Problem (Master Problem) and three interdependent sub-problems one of which is the Price Determination Sub-Problem. In the Master Problem "EUPHEMIA searches among the set of solutions for a good selection of block and MIC orders that maximises the social welfare. Once an integer solution has been found for this problem, EUPHEMIA moves on to determine the market clearing prices." i.e. the Price Determination Sub-Problem.

From the above, it is clear that viewing the algorithm solely as the intersection of bid and ask curves, while intuitive, oversimplifies its underlying mechanics. The algorithm has undergone and continues to undergo incremental change [35,36], making it a dynamic rather than static construct. In contrast to EUPHEMIA, for which only a high-level description is publicly available, the algorithm governing the Irish Balancing market for example is fully described in [37]6.

2.3. Speculators

We explain the concept of speculators in the Irish electricity market using a simple example. Consider a single speculator and delivery period. Suppose the speculator forecasts prices of €X/MWh in the DA/IDA1/IDA2/IDA3 ex-ante markets and forecasts a price of €(X+20)/MWh in the Balancing market. Based on these forecasts the speculator’s strategy is to buy in the cheaper ex-ante markets and implicitly sell in the more expensive balancing market (given that a speculator does not consume or generate electrical energy its balancing market position is simply the sum of its ex-ante positions for that trading period multiplied by a factor of $-1$). If they buy 50MW/25MW/0MW/0MW in the DA/IDA1/IDA2/IDA3 markets, then their balancing market position will be a sell of 75MW. If the speculator’s forecasts are correct they stand to make a profit of €750 because the cost of the buys, €${\displaystyle \frac{\left(50X+25X+0X+0X\right)}{2}}$, is less than the income from the sell, €${\displaystyle \frac{75\left(X+20\right)}{2}}$. Note the ${\displaystyle \frac{1}{2}}$ factor in the calculation is required because we are dealing with 30min intervals. If the forecast is incorrect then they could make more or less than €750, including a loss.

3. Materials and Methods

In this section we describe the underlying datasets and the empirical analyses. For context, we note that the data used in the study coincides with a diverse and volatile set of energy price regimes including the Covid-19 pandemic7 and the European energy crisis. Any reference to accompanying material in the appendices is for replicability and transparency purposes with the key details being captured in this Section 3.

3.1. Datasets

SEMOpx and SEMO are two of the bodies involved in the operation of the Irish electricity market [41,42]. Some of the datasets they publish include:

• Granular Participant Data: For each of the four ex-ante markets (Section 2.1) SEMOpx publishes a distinct file called the ETS Bid File. We use these files to interrogate the order and trade quantity data at a participant and trading period level of granularity. In each ETS Bid File positive (negative) order quantities represent purchase (sell) orders; the same convention applies to trade quantities. The DA ETS Bid file for example is published on a day+1 basis relative to the trading day and typically it consists of in excess of circa twenty thousand rows with an average of three hundred plus participants per auction.
• Bid Ask Curve Data: For each ex-ante auction SEMOpx publish a BidAskCurve file containing a monotonic increasing (decreasing) and anonymised view of the sell (buy) orders for each trading period in the auction8.
• Other Data:

  –

  PUB_MnlyRegisteredCapacity files which provide participant registration data such as registered plant capacity and FuelType (if applicable).

  –

  PUB_30MinImbalCost file containing the Balancing Market price, $P_i^B$, for trading period i. Similarly, MarketResult files containing DA, IDA1, IDA2 and IDA3 market prices (i.e. $P_i^{\mathrm{DA}}$, $P_i^{\mathrm{IDA}1}$, $P_i^{\mathrm{IDA}2}$, and $P_i^{\mathrm{IDA}3}$).

In the Irish electricity market, participants may register one or more unit types, including Supply, Generator, Assetless, and Trading units, among others. These units have flexibility in their behaviour, so we follow the sequence of steps outlined in Appendix B to identify speculators.

3.2. Market and Speculator Analysis

Quantities

The DA market accounts for approximately 85% of the total volume traded across the DA, IDA1, IDA2, IDA3, and IDC markets. Therefore, we examine how DA volumes have evolved for speculators and for the market as a whole, with the latter providing context for the former. Using information in the DA market ETS Bid Files we derive a per trading period time series that shows

• The quantities of energy that participants were willing to buy or sell, we refer to these as the order quantities.
• The quantities of energy bought or sold which we term the matched quantities.

Marginal Participants

To understand which participants are marginal and whether or not this pattern has changed over time for each trading period (see Appendix D for details):

• We retrieve the bid and ask curves from the relevant BidAskCurve file and determine how they intersect.
• The intersection point(s) are then cross-referenced against the participant order data in the relevant ETS Bid File to determine which unit, or units, are marginal.

A by-product of the marginal analysis is the identification of bid and ask curve intersections. As a result, in addition to identifying marginal participants we also examine how DA bid and ask curves intersect and how these intersection patterns have evolved over time. Given the stepwise nature of the bid and ask curves in the Irish electricity market there are three possible types of intersection: horizontal (the curves are horizontally parallel and intersect over a ranges of quantities), vertical (the curves are vertically parallel and intersect over a ranges of prices) and perpendicular (the curves intersect perpendicularly at a single price and quantity point). For a perpendicular intersection, if at the point of intersection the ask curve is horizontal, ACH, we denote it as a perpendicular (ACH) intersection. Similarly, if at the point of intersection the bid curve is horizontal, BCH, we denote it as a perpendicular (BCH) intersection. In Appendix E we provide examples of different bid ask curve intersection types.

Aggregate speculator behaviour and profitability

To understand how aggregate speculator behaviours have evolved over time we start by examining the speculator DA order (i.e. pre-market clearing) quantities referenced in Section 3.2.0.1. It is tempting to compare the speculator order quantity distributions (e.g. using kernel density estimates) at discrete points in time. However, here we take a less complicated approach yet still manage to capture the high level trends in a succinct manner. For trading period i, we denote the DA market price as $P_i^{\mathrm{DA}}$ and the the standard deviation of DA market prices for the corresponding hour over the previous 20 days as ${\mathrm{SD}}_i$. We then partition the price domain into three distinct intervals defined as follows:

• Interval 1: $[-500,P_i^{\mathrm{DA}}-{\mathrm{SD}}_i]$
• Interval 2: $[P_i^{\mathrm{DA}}-{\mathrm{SD}}_i,P_i^{\mathrm{DA}}+{\mathrm{SD}}_i]$
• Interval 3: $[P_i^{\mathrm{DA}}+{\mathrm{SD}}_i,3000]$

Here the $-500$ and 3000 values correspond to the minimum and maximum permissible DA market prices9. For each price interval we calculate the sum of the speculator sell, and separately the speculator buy, order quantities that fall within that bucket. We then aggregate the values for each price interval by month and plot the monthly totals from November 2018 to December 2022.

Next, we consider the aggregate matched (i.e. cleared) speculator quantities. While we are primarily interested in the aggregate DA matched speculator quantities, motivated by the structural market change referenced in Section 2.1.0.1 and the hypothetical example in Section 2.3, we also examine the corresponding IDA1 and Balancing market values. With Speculator denoting the set of speculators participants and $Q_{i,j}^m$ representing the quantity of power bought or sold (i.e. matched quantity) in trading period i by speculator j in market m we define

$$\mathrm{Speculator}\mathrm{DA}\mathrm{Matched}{\mathrm{Quantity}}_i=\sum _{j\in \mathrm{Speculator}}{Q}_{i,j}^{\mathrm{DA}}$$

(1)

$$\mathrm{Speculator}\mathrm{IDA}1\mathrm{Matched}{\mathrm{Quantity}}_i=\sum _{j\in \mathrm{Speculator}}{Q}_{i,j}^{\mathrm{IDA}1}$$

(2)

$$\mathrm{Speculator}\mathrm{Balancing}\mathrm{Market}{\mathrm{Quantity}}_i=\sum _{j\in \mathrm{Speculator}}{Q}_{i,j}^B$$

(3)

We calculate these variables for each trading period over the study horizon and present pair plots in order to identify any similarities or differences in aggregate speculator behaviours before and after the structural market change. For brevity we omit pair plots of the corresponding IDA2 and IDA3 Speculator values as these markets account for only a small share of ex-ante trading volumes. For clarity, similar to the hypothetical speculator example in Section 2.3, with $E=\{\mathrm{DA},\mathrm{IDA}1,\mathrm{IDA}2,\mathrm{IDA}3\}$, the $Q_{i,j}^B$ values in equation 3 are obtained via

$$Q_{i,j}^B=-\sum _{m\in E}{Q}_{i,j}^m$$

(4)

Finally, ignoring trading or transaction costs which in the Irish electricity market are minimal, our approach to estimating speculator profitability mimics the earlier hypothetical example. With $A=\{\mathrm{DA},\mathrm{IDA}1,\mathrm{IDA}2,\mathrm{IDA}3,\mathrm{B}\}$, Speculator j’s profit and loss in trading period i is then estimated via

$$-{\displaystyle \frac{1}{2}}\sum _{m\in A}{P}_i^m{Q}_{i,j}^m$$

(5)

Here $P_i^m$ represents the market price for trading period i in market m. The $-{\displaystyle \frac{1}{2}}$ factor in equation 5 reflects the buy and sell sign convention and 30 minute delivery periods. This calculation is then repeated for all speculator participants across all trading periods in the study horizon to produce an aggregate estimate of speculator profitability

3.3. Price Inertia Analysis

Our approach to measuring price inertia in the Irish DA electricity market involves applying a small horizontal shift to either the bid or ask curve and calculating the resulting intersection price. The difference between the market price and this new intersection price serves as a measure of price inertia. One could argue that obtaining the resulting intersection price would require a full rerun of the EUPHEMIA algorithm. However, since we apply only small horizontal shifts of 1 MW or 10 MW, minor relative to average forecast demand in the Irish DA market which was approximately 4300 MW per trading period over the study horizon, and the algorithm is not publicly accessible, we adopt a reasonable simplifying assumption: that a full rerun of the market pricing algorithm is not required. This approach is consistent with the existing literature where studies examining supply and demand curves in EUPHEMIA related contexts similarly rely on approximations or simplifications due to the unavailability of the algorithm. The process of applying a small horizontal shift of XMW (with X equating to 1MW or 10MW) to the DA ask curve in trading period i is as follows:

• Retrieve the bid and ask curve for trading period i from the relevant DA BidAskCurve file.
• Add the quantity X MW to each $(quantity,price)$ point on the ask curve
• Determine where this horizontally shifted ask curve intersects the bid curve. We call the resulting price the Simulated DA market price for trading period i, denoted by $P_i^{\mathrm{SDA}}$.
• With $P_i^{\mathrm{DA}}$ and ${\mathrm{SD}}_i$ as per Section 3.2.0.3, we define the Price Difference and Custom Metric values in trading period i as

  $$\mathrm{Price}{\mathrm{Difference}}_i=P_i^{\mathrm{SDA}}-P_i^{\mathrm{DA}}$$

  (6)

  $$\mathrm{Custom}{\mathrm{Metric}}_i={\displaystyle \frac{\left|\mathrm{Price}{\mathrm{Difference}}_i\right|}{S{D}_i}}$$

  (7)

If we consider the Price Difference time series that results from equation 6 on a stand-alone basis it is potentially misleading given the different commodity price regimes that existed over the study horizon (Figure A9 Appendix N presents a plot of the corresponding electricity market prices). Hence our interest in contextualising or normalising the Price Difference metric; we could do so by dividing by $P_i^{\mathrm{DA}}$ but this is undefined in trading periods in which the DA price is €0/MWh. Our approach is to instead divide by the standard deviation of market prices for that hour over the previous twenty days (i.e. ${\mathrm{SD}}_i$); other approaches are possible. We present the results of small horizontal shifts in the ask curves, results for small horizontal shifts in the bid curves are available but are omitted due to space constraints. Next, we briefly discuss two price inertia approaches that build on the above concepts and further enhance our intuition and understanding (we note however that in the following context the assumption of a small shift may not always hold true). The details are as follows.

Price Inertia Distribution

• For each trading period adjust the ask curve in $+1$MW increments until the simulated market price, $P_i^{\mathrm{SDA}}$, differs from the published market price. Let ${\mathrm{Shift}}_i^{+1\mathrm{MW}}$ denote the shift required to observe the price change where the $+1\mathrm{MW}$ superscript indicates that we have used $+1$MW increments. Use equation 6 to calculate the associated price difference which we denote as $\mathrm{Price}{\mathrm{Difference}}_i^{+1\mathrm{MW}}$.
• Repeat the previous step except use $-1$MW decrements to derive the corresponding ${\mathrm{Shift}}_i^{-1\mathrm{MW}}$ and $\mathrm{Price}{\mathrm{Difference}}_i^{-1\mathrm{MW}}$ values.
• For each trading period, define and calculate the Min Shift and Price Impact via

  $$\begin{array}{cc}\hfill \mathrm{Min}{\mathrm{Shift}}_i& =min\left(|{\mathrm{Shift}}_i^{+1\mathrm{MW}}|,|{\mathrm{Shift}}_i^{-1\mathrm{MW}}|\right)\hfill \\ \hfill \mathrm{Price}{\mathrm{Impact}}_i& =\left\{\begin{array}{cc}|\mathrm{Price}{\mathrm{Difference}}_i^{+1\mathrm{MW}}|,\hfill & \mathrm{if}\mathrm{Min}{\mathrm{Shift}}_i=\left|{\mathrm{Shift}}_i^{+1\mathrm{MW}}\right|\hfill \\ |\mathrm{Price}{\mathrm{Difference}}_i^{-1\mathrm{MW}}|,\hfill & \mathrm{if}\mathrm{Min}{\mathrm{Shift}}_i=\left|{\mathrm{Shift}}_i^{-1\mathrm{MW}}\right|\hfill \end{array}\right.\hfill \end{array}$$

  (8)

In Section 4.2 Figure 10 we present density histograms of the Min Shift and Price Impact values.

Order Example

This approach involves removing the order data for a single (speculator) participant, reconstructing the bid and ask curves using the remaining participants’ order data, and determining the new intersection price. Implementation details for this alternative approach and related examples are provided in Appendix L.

4. Results

4.1. Market and Speculator Analysis

Quantities

Figure 1 presents stacked area charts showing the sell order and sell matched quantities (Section 3.2.0.1) in TWh, grouped by FuelType10 and month for the Irish electricity market. The dashed vertical black line, also present in a number of subsequent plots, highlights the point in time at which the structural market change occurred (Section 2.1.0.1), recall the sign convention that sells (buys) are represented by negative (positive) quantities. It can be seen that approximately 52% of the sell order quantities will on average be matched. For conciseness, we do not present the corresponding buy plots, but it is worth noting that on average 83% of buy orders are matched. The presence of the Coal, Oil and Distillate categories in the sell order quantities plot and their absence for significant periods of time in the sell matched quantities plot is indicative of the Irish electricity market DA merit order.

A change in speculator sell order and sell matched quantities (i.e. red time series) post the structural market change is evident in Figure 1. From the underlying data it is seen that on average pre (post) the structural market change, speculators accounted for circa 10% (20%) of DA order quantities and approximately 2% (6%) of DA matched quantities. We note that while we identified 50 speculators in the Irish DA electricity market prior to the structural market change an average of 16 (14) speculators would submit sell (buy) orders to the DA whereas post the structural market change the corresponding number is 30 (24) speculators i.e. almost a doubling in the number of active speculators. For context we note that over the November 2018 to December 2022 timeframe there were 472 distinct participants in the Irish electricity DA market. Separately, it is interesting to note that for the IDA1 market which accounts for approximately 10% of traded volumes, speculators units have a significantly larger market share. On average speculators account for 54% (42%) of buy (sell) IDA1 order quantities and 46% (33%) of buy (sell) IDA1 matched quantities.

Marginal Participants

The top plot in Figure 2 presents a stacked area chart showing the percentage of marginal observations, grouped by fuel type and month. It can be seen that gas and multi-fuel units have exhibited a declining tendency to act as the marginal unit; prior to the structural market change this cohort was marginal in approximately 35% of trading periods compared to around 20% afterward. In contrast, speculator participants appear, at least visually, to be marginal in a greater proportion of trading periods post-change (rising from 49% before to 66% after). That speculators are marginal in the majority of trading periods stands in contrast to their overall minority presence in the market, as outlined in the previous paragraph.

The bottom plot in in Figure 2 presents a stacked area chart showing how the DA bid and ask curve intersections (i.e. vertical, horizontal or one of the two types of perpendicular intersection) have evolved since the market inception. A noticeable change occurs post the structural market change.

Aggregate Speculator behaviour and profitability

For insights into aggregate speculator pre market clearing behaviours we take summed DA speculator sell order quantity data from the top plot in Figure 1 (red time series) and split it by price interval (Section 3.2.0.3). Doing so, we arrive at the top plot in Figure 3. Again, to provide context, if we repeat the process but do it for non-speculators we obtain the bottom plot in Figure 3. For speculators it is clear that a lot of the increase in sell order quantities post the structural market change is associated with the $[P_i^{DA}-S{D}_i,P_i^{DA}+S{D}_i]$ price interval (orange time series). Examining the corresponding non-speculator data there is no obvious step change.

Plots of buy order quantity data, split by price interval for speculators and non-speculators, are provided in Appendix I, Figure A3. Similar patterns emerge from these plots, albeit there is a slight, visually imperceptible uptick in non-speculator buy order quantities within the $[P^{DA}i-S{D}_i,P^{DA}i+S{D}_i]$ price interval following the structural market change. Separately, we note that on average 97% of non-speculator buy order volumes occur in the $[3000,P_i^{DA}+S{D}_i]$ interval which is an indication of the inelasticity of demand.

Figure 4 presents scatter plots of the Speculator IDA1 Matched Quantity (equation 2) against the corresponding Speculator DA Matched Quantity (equation 1) distinguishing between observations pre and post the structural market change. It is seen that post the structural market change the magnitude of speculator DA positions increased; a negative correlation between speculator positions in the DA and IDA1 markets is also clear (i.e. it is often the case that speculator DA matched quantities are partially or fully unwound in IDA1). In a similar manner Figure 5 plots the Speculator Balancing Market Quantity (equation 3) against the corresponding Speculator DA Matched Quantity, again we distinguish between observations pre and post the structural market change. It is seen that pre the structural market change there was a negative correlation between between the time series, but post the change the negative correlation is not apparent.

To provide some insight into aggregate speculator behaviour post January 2021 and also to highlight the general dynamism in their commercial behaviours, Figure 6 presents a granular time series view of two weeks worth of speculator data. The top plot in Figure 6 shows Speculator DA Matched Quantity in green and Speculator IDA1 Matched Quantity multiplied by a factor of $-1$ in brown, the latter is denoted as Speculator IDA1 Matched Quantity (inverted). As per earlier comments, the visible correlation between the time series illustrates that speculator DA matched quantity positions are frequently, but not always, offset in the IDA1 market. The middle plot in Figure 6 shows Speculator Balancing Market Quantity values; given the preceding commentary (i.e. speculator DA positions are often partially or fully unwound in the IDA1) it is understandable that the magnitude of Speculator Balancing market quantities is less than the magnitude of corresponding speculator DA and IDA1 values. Taking the Speculator DA Matched Quantity time series in green from the top plot but splitting it into speculator buy matched quantities and speculator sell matched quantities, we arrive at the bottom plot. This plot highlights how speculators can have differing behaviours, that is speculators are not a homogeneous population. Moreover, this plot, and the other plots in Figure 6 once again illustrate the dynamic non-stationary commercial behaviours of speculators in the Irish electricity market.

Using equations 4 and 5, an estimate of the aggregate speculator profit and loss (P&L) from November 2018 to December 2022 is €$76.8$ million11. Of the total amount €$66.2$ million is attributable to calendar years 2021 and 2022. The six speculators referenced in Section 4.1.0.2 account for €$32.4$ million (i.e. 42% of estimated speculator profitability). Of the €$76.8$ million total, the amount that relates to speculators taking opposing positions in ex-ante markets (see Appendix G) is estimated at €$29.9$ million. Examining the granular P&L data (i.e. by speculator and trading period) then it can be seen that in circa 54% (46%) of observations in which there was a cash-flow it was positive (negative). This is indicative of the challenge speculators face in pursuing profitable trading strategies i.e. speculation in short-term electricity markets is not a risk-free activity.

4.2. Price Inertia Analysis

Taking bid ask curves from the November 2018 to December 2022 time frame, if we apply a +10MW horizontal shift to the ask curve and calculate the percentage of trading periods per day that have a Price Difference (equation 6) of €0/MWh, then we arrive at Figure 7. A step change in the percentage of zero price difference observations per day post the structural market change is visible. Using the time series values from the +10MW horizontal shift to the ask curve, the top plot in Figure 8 displays the Price Difference values per trading period while the bottom plot presents the corresponding Custom Metric values (equation 7). The Custom Metric plot shows a reduction in the horizontal shift impact from 2021 onwards. Examining the underlying data it is observed that in approximately 72% (36%) of all trading periods pre (post) 1st January 2021 the application of a +10MW horizontal shift would lead to a change in the market price. For those trading periods where the market price would change, it would lead to an average reduction of 3.6% (0.8%) in the DA price pre (post) 1st January 2021. We note in passing that, when comparing the proportion of trading periods with non-zero price differences before and after 1 January 2021, a two-proportion z-test rejects the null hypothesis that these proportions are the same12. In Figure 9 we take the Price Difference and Custom Metric time series values from Figure 8 and plot them against the corresponding DA price. From the scatter plots the difference in distributions pre and post the structural market change is evident. Note that in Appendix J Figure A4 and Figure A5 present histograms of the time series observations presented in Figure 8 while in Appendix K we present the corresponding +1MW price inertia analysis.

Figure 10 presents the price inertia distribution analysis described in Section 3.3. The density histogram on the left compares the distributions of the minimum shift required to trigger a market price change before and after 1 January 2021. It shows that, prior to this date, small shifts in the ask curve were more likely to result in a price change than comparable shifts after the change. Finally, in Appendix L we take a single speculator and simulate its impact on market prices. This approach is conceptually similar to the price inertia method although the assumption of a small shift in the demand or supply curve may not always hold.

Figure 10. Price Inertia Distribution. Plot on the left is a histogram of the minimum absolute shift required to trigger a market price change (note: x-axis range restricted for display purposes), plot on the right is the estimated price impact.

Figure 10. Price Inertia Distribution. Plot on the left is a histogram of the minimum absolute shift required to trigger a market price change (note: x-axis range restricted for display purposes), plot on the right is the estimated price impact.

5. Discussion

It is perhaps unsurprising that the Irish electricity market exhibits evidence of both the dynamism in aggregate speculator behaviours as well as a non-homogeneity in their behaviours. The empirical analysis also reveals that following the structural market change speculators participation in the DA market underwent a step change. In particular, their buy and sell order volumes within the $[P_i^{\mathrm{DA}}-{\mathrm{SD}}_i,P_i^{\mathrm{DA}}+{\mathrm{SD}}_i]$ price interval increased. At first glance, the frequency with which speculators were marginal, exceeding their overall market share both before and after the structural market change, may seem counterintuitive. However, on further reflection, this outcome is not entirely unexpected as the strategies employed by certain speculators involve placing a portion of their orders near the market-clearing price.

The price inertia analysis offered a straightforward and intuitive methodology. It was notable that the percentage of trading periods in which a small shift in the ask curve would have resulted in a market price change was relatively high and, in some cases, the resulting price changes were substantial. These findings highlight the susceptibility of DA electricity market prices to jumps or shocks, a characteristic feature with which many market participants will already be well versed. While our analysis focuses on the DA setting, the underlying concept is analogous to how participants often assess market depth and price sensitivity in other market environments. The shift in price inertia levels before and after the structural market change also stood out.

A plausible hypothesis linking the two preceding observations is as follows: after the structural market change, speculators began providing DA liquidity that had previously been supplied via EUPHEMIA’s interconnector scheduling between the Irish and British electricity markets. This behaviour aligns with one of the motivations for Virtual Bidding noted by [14] namely enhanced market liquidity and price transparency. A significant portion of this additional speculator liquidity appeared near the market-clearing price, potentially contributing to the observed increase in price inertia levels. While the correlation is suggestive, it does not establish causation. The ability to rigorously test this hypothesis is constrained by two key challenges:

• Pricing algorithm access: the market-clearing process is complex and only partially disclosed. Without full access to the EUPHEMIA algorithm, it is difficult to run counterfactual scenarios or isolate behavioural effects. We note [13] make related observations in their treatment of Virtual Bidding.
• Market dynamics: As noted by [15], market outcomes are shaped by evolving participant behaviours. Even subtle shifts in non-speculator actions either individually or in aggregate, for example the slight increase in non-speculator buy order volumes discussed in Section 4.1.0.3, could contribute to the observed changes.

Further research is warranted to explore this hypothesis more formally. One possible avenue is to forecast price inertia levels and identify the most influential explanatory variables, if indeed a model with meaningful predictive power can be developed. Even without additional research the empirical analysis hints at limitations in applying fundamental electricity price forecasting approaches to short-term markets. Speculators exhibit dynamic and adaptive behaviours, making the calibration of related input assumptions inherently challenging. Moreover, the price inertia analysis shows that small changes in supply or demand can, in certain trading periods, lead to material price movements, nonlinear effects that may be difficult to capture using fundamental models (or indeed many other modelling approaches).

6. Conclusions

In this paper we presented two distinct empirical analyses on the Irish electricity market. The study setting was representative of typical short-term European electricity markets with a structure that ensured a non-trivial degree of liquidity and the environment featured a high variable renewable generation penetration. The study period also coincided with a diverse range of commodity pricing regimes. The first piece of analysis examined the presence and strategic behaviour of speculators while the second addressed market price inertia. For speculators it was observed that the proportion of trading periods in which they were marginal was significantly in excess of their traded market share, they exhibited increased activity post a structural market change and the dynamism and non-homogeneity in their behaviours was also evident. The price inertia analysis introduced an intuitive and easily understood methodology and it also underlined the susceptibility or sensitivity of short-term electricity market prices to small changes in demand or supply. While an increase in the market price inertia levels post the structural market change was observed we could only conjecture as to the underlying causes given the domain complexity (i.e. the dimensionality of the repeated auction mechanism and lack of access to the market pricing algorithm). This research has two key implications. First, the findings on speculators and price inertia suggest inherent challenges in the ability of short-term fundamental market modelling approaches to accurately forecast electricity prices. Second, the price inertia methodology provides an additional metric that could be valuable when evaluating structural market change proposals. Our analysis contributes to the existing literature by reinforcing the understanding that short-term electricity markets are complex non-stationary environments with numerous interacting factors.