26Power, Commodities and Betting Connectivity

Europe’s cross-border intraday power market matches orders from twenty-five countries through one shared order book, and allocates the interconnectors’ capacity with each trade, first come, first served. When capacity across a border is scarce, the last megawatt goes to whoever reaches the book first. In the chapter’s model a participant $10\,\mathrm{m}\mathrm{s}$ from the book, against a rival at five, obtains the last megawatt in 4.5% of the contests; at $2\,\mathrm{m}\mathrm{s}$, in 98.5%.

Book 3 described intraday power, market coupling, bidding zones, balancing and schedule nominations, and the betting exchanges. This chapter is about how firms connect to them: the shared order book and the local trading systems in front of it, the grid operators’ scheduling systems, the commodity exchanges of chapter 11’s map, and a betting exchange’s interfaces and charges. The capacity race is a labelled simulation; the rules are the operators’ own.

26.1 Intraday power interfaces and the shared order book

The single intraday coupling (SIDC) runs one shared order book, a capacity management module holding the cross-zonal capacities the grid operators release, and a shipping module (Box 26.1). A participant does not connect to the shared book directly: it trades through its market operator’s local trading system, which passes orders to the book and shows it the orders of other operators’ participants where capacity allows. Within a price level, orders rank by the time they were registered; across a border, a match is possible only while capacity remains, and each match consumes it.

A scarce border is a race. Imagine a price difference opening between two zones with room left on the interconnector for one more order: every participant who sees it sends an order, and the first to reach the shared book gets the capacity. The chapter’s model has five rivals whose latencies to the book are lognormal around 5, 10, 20, 40 and $80\,\mathrm{m}\mathrm{s}$, with a spread of 0.3, and varies the firm’s own median (all assumptions: where the book is hosted and how fast each local system is are not published).

def capacity_race(participants, slots=1, n=40000, seed=0):
    """Each event: every participant sends one order; the first `slots` arrivals get the scarce capacity."""
    rng = np.random.default_rng(seed)
    t = np.column_stack([p.median_ms * np.exp(p.sigma * rng.standard_normal(n)) for p in participants])
    rank = np.argsort(np.argsort(t, axis=1), axis=1)
    got = rank < slots
    return {p.name: float(got[:, i].mean()) for i, p in enumerate(participants)}

The value follows. If the price difference the capacity captures averages EUR 12 per megawatt-hour (an assumption), a firm at $2\,\mathrm{m}\mathrm{s}$ expects EUR 11.82 per megawatt-hour it offers into such contests and a firm at $10\,\mathrm{m}\mathrm{s}$ EUR 0.54: the whole value of a scarce border goes to the fastest participant, and the second fastest gets almost nothing. With three orders’ capacity left, the same $10\,\mathrm{m}\mathrm{s}$ firm gets 95.4% of contests. Scarcity, not the market, turns an intraday power desk into a latency business, and only on the borders and hours where capacity binds.

The gate closure bounds the race in time: cross-zonal trading for a delivery period stops 60 minutes before it (30 after the planned change), and during the intraday auctions the continuous market’s cross-zonal allocation stops on the auctioned borders. A desk’s clock therefore has three kinds of moment: continuous trading with capacity, auction windows without it, and the gate.

26.2 Grid-operator scheduling systems

After trading, the positions become schedules: a balancing responsible party nominates to its grid operator the energy it will inject, withdraw and exchange, period by period, before the operator’s deadlines. These exchanges are not trading messages but documents, sent over ENTSO-E’s Energy Communication Platform, which offers reliable, secure delivery over AMQP, the MADES web service (IEC 62325-504) or shared folders. Latency here is minutes, not milliseconds; what matters is that every trade has a schedule, that the documents are acknowledged, and that a failed delivery is noticed before the deadline, the same reconciliation discipline as a drop copy (chapter 23), on a slower clock.

26.3 Commodity-exchange connectivity

The commodity futures exchanges are exchanges like the others, on chapter 11’s map: ICE Futures Europe’s matching in Basildon, the European energy and metals contracts on the same continental campuses as the financial ones. Their connectivity is chapter 9’s: cross-connects, colocation, feeds and gateways, with the fairness questions of chapter 23. What differs for a power or gas desk is the combination: a futures position on an exchange, an intraday position through a local trading system and the shared book, and schedules to a grid operator, three connections on three clocks for one physical exposure.

26.4 Betting-exchange interfaces and their data charges

A betting exchange sells access to its order books through an application programming interface, and prices it by what the client asks of it. Betfair’s documentation is explicit (Box 26.2): data requests are limited by weight, 200 points per request, where each market in a request costs the weight of the projection asked for; transactions above a threshold per hour are charged; and the exchange recommends its streaming interface over polling.

def markets_per_request(weight, limit=200):
    return limit // weight


def poll_requests_per_s(markets, weight, hz, limit=200):
    return math.ceil(markets / markets_per_request(weight, limit)) * hz


def poll_staleness_ms(hz, rtt_ms):
    """Polling every 1/hz seconds: on average half an interval old, plus the request's round trip."""
    return 1000.0 / (2 * hz) + rtt_ms


def charged_transactions(per_hour, threshold=5000):
    return max(0, per_hour - threshold)

Polling 1 000 markets five times a second for best offers takes 40 markets a request, 25 requests a poll, 125 requests a second; for the full ladder with traded volumes, 6 markets a request and 835 requests a second. Either way the data is on average $110\,\mathrm{m}\mathrm{s}$ old with a $10\,\mathrm{m}\mathrm{s}$ round trip, half the polling interval plus the trip. A stream sends each change once, when it happens. And a strategy re-quoting 30 markets four times a minute sends 7 200 transactions an hour, 2 200 above the threshold: on a venue that charges by the action, the cancel-and-replace habits of an exchange market maker are a cost line.

26.5 Tutorial: the last megawatt

26.6 Build: the power and betting connectivity model

26.7 Exercises

26.8 Problem: The Last Megawatt Across the Border

26.9 Interview questions