Understanding the “arbitrage” value of a large battery in SA

Print Friendly


Given recent events in South Australia’s energy market and the recent’s tweet by Elon Musk of “100 days or it’s free…” to deliver 100MWh’s of energy storage, there has been much commentary on the value of energy storage.  I thought it instructive for RenewEconomy readers to delve into the economics of one of the key values of energy storage, “arbitrage”.

As a former energy trader, I would explain this concept simply as “buy low, sell high” or alternatively in the energy spot market it is the concept of purchasing and storing energy at low price times only to generate and sell into the market at high priced times.

Whilst conceptually simple, the purchase price of a battery is high, there are physical limitations on charge/discharge and the frequency of high spot prices is relatively infrequent.  The following calculations endeavour to establish a benchmark of the historic economics in South Australia.

To value the arbitrage economics of energy storage I have taken South Australian spot market prices from 1999 to the end of February 2017 and then assumed that we had perfect foresight, i.e. if we had 100% ability each day to charge at the lowest priced periods and then discharge at the highest priced periods (a bold but unrealisable goal).

I then assumed that our round-trip efficiency was 80% (i.e. what we get out compared against what we put in) which is typical for a large lithium-ion energy storage system.

Then, I analysed what I believe is missing in most of the analyses I have seen, which is consideration of how long the battery takes to charge and discharge i.e. most people have assumed simply taking the highest and lowest prices each day to estimate the arbitrage value, but if your battery takes 2 hours to charge and discharge then the total price you receive will be noticeably lower as this means it is the difference between the average of each the four highest  and four lowest prices each day.  We can then express this as algebraically for a charge/discharge time as:

Revenue($/MWh) = Round trip efficiency (%) X (S(P1+..Pn)/n –  S(P(48-n) +..P48)/n)

Where n is the number of half hours of storage at full power given that the market settles as 48 half-hours per day and that P1 is the highest price, P2 is the second highest price, and so on until we get to P48 which is the lowest price.  If we do this calculation every day for a year these are the results we have are in the following table.

Table 1 – SA Average daily return $/MWh per MWh each year for given charge/discharge durations

Year Days 4 hours 2 hours 1 hour 30 minutes
1999 365 $100.40 $137.42 $175.33 $204.55
2000 366 $102.63 $138.39 $181.25 $223.47
2001 365 $64.58 $90.00 $114.90 $142.16
2002 365 $51.84 $82.99 $130.45 $164.49
2003 365 $28.87 $40.47 $55.30 $70.07
2004 366 $63.91 $95.87 $131.94 $162.55
2005 365 $43.99 $65.60 $90.97 $123.00
2006 365 $57.39 $88.40 $130.66 $169.07
2007 365 $65.62 $85.41 $106.54 $125.58
2008 366 $160.72 $220.34 $251.29 $272.18
2009 365 $161.98 $232.78 $266.47 $289.57
2010 365 $90.32 $130.03 $172.59 $198.01
2011 365 $55.09 $69.31 $86.48 $103.62
2012 366 $33.06 $47.31 $65.56 $81.17
2013 365 $74.82 $110.95 $159.82 $223.36
2014 365 $41.97 $63.24 $90.94 $123.72
2015 365 $72.92 $115.04 $181.50 $265.48
2016 366 $142.62 $203.95 $272.54 $358.10
2017 59 $269.38 $414.70 $500.49 $570.15
Whole Year Average $78.49 $112.08 $148.03 $183.34

What this shows is that if we had a battery that cycled daily, this was the return per MWh of energy storage we would expect each day for a given duration of charge/discharge cycle.  This duration is also measured alternatively as C rating, where this value is equal to 1/duration (hours) i.e. a 30-minute battery is equal to 1/0.5 = 2.0 C and a 2-hour battery is equal to 1/2 or 0.5 C.

What is important to note is that the shorter duration charge cycle or higher C rating battery has a much higher return per MWh of storage because the highest prices only occur for a single period, so extending the discharge time over lower priced periods reduces the average price received.

Whilst a 2.0 C battery will always have better financial performance than a lower C rating battery, it is often more expensive per MWh of storage.  The question then becomes whether a battery system with a lower C rating can be a better choice if its cost per MWh is sufficiently low.

Another way of considering it is that if we look at annual averages, a 2.0 C battery could break even at $183.34/MWh, whilst a 0.25 C battery would need to be as low as $78.49/MWh, and a 0.5 C battery could break even at $112/MWh. This means that whilst Tesla has hit the news with a compelling price offer, if a 2 C battery could be installed at a price 50% higher than Tesla, it would potentially be a better economic choice when considering arbitrage alone.


If we take the example that Elon Musk Tweeted, he talks of $250/kWh of capacity at the Powerpack level but the installed cost will likely be close to double this so around $500,000/MWh installed.

With an expected life of 3650 daily cycles over 10 years and if we ignore the fact that the battery will degrade to as little as 60% of its original capacity and discounted cash-flow analysis, but still considering 80% round trip efficiency, we get an effective price of $500,000/3650/0.8, which is $171/MWh.

The Tesla technology at present is 0.5C which could have returned $203.95/MWh in 2016 but would have averaged only $112.08/MWh in the past, which is insufficient to recover the investment costs at today’s prices.

What this means is that the potential investment is unlikely to make sense at grid scale on pure arbitrage value alone for a battery just yet, but it is getting much closer as the economics are approaching the level of an idealised arbitrageur.

A battery of 100MW would also have a material impact on spot prices in South Australia, such that it would marginally increase the lower prices and potentially reduce some of the peak prices quite significantly, thereby reducing the financial performance of the battery even further.

There are other income streams such as the Frequency Control Ancillary Services (FCAS) market but given that this is a balancing market to respond quickly to differences between dispatched power and system demand, it is not usual that such large MW volumes would be necessary most of the time and hence it may be considered oversized plus FCAS is even more volatile than energy spot markets.

Batteries can also perform “black start recovery”, i.e. return a system from a blackout condition which is an additional source of revenue but unlikely to be politically popular if it is called upon.

However, the operation of a grid connected battery is usually such that providing multiple services often conflict with each other meaning that the total value will be less than the sum of the potential maximum value of all the individual services.  i.e. given the payback on arbitrage which is likely to be the largest value is marginal and the uncertainty is high, a commercial investor or battery supplier is not likely to fund a battery wholly themselves and will require “co-investment” or a subsidy before a battery is installed.

This is obviously the case in South Australia, so careful examination of the economics of the battery including both the power (MW) and energy (MWh) requirements will be needed to ensure that the people of South Australia get the right sized solution at the right price.

Warwick Forster is managing director of Apogee Energy



Share this:

  • Craig Memery

    Very nice analysis Warwick,
    Of course another of the key assumptions that comes with perfect forsight is anticipating events that are caused not by predictable high demand but by less predictable transmission constraints and generator outages.
    The idea of whether batteries for black start is an interesting one I’ve been pondering. On one hand, I imagine it would add to the inverter cost (this is why for example Murraylink DC connector into South Australia can’t provide restart services) On the other, system restart ancillary services are worth about $100m a year across the NEM even if they are not used (and the ones procured in South Australia proved ineffective in September – money for nothing) so the inverter upgrade might be worth it the battery could access some of SRAS payment. But could the state of charge when the system restart is needed be relied on?

    • Warwick Forster

      Craig, you’re raising the valid point of potential conflicts between each of the value streams. To maximise the arbitrage value, you need to fully charge and discharge the battery, meaning significant periods during the day when the battery cannot inject or withdraw energy due to its 0 or 100% states of charge respectively. Never mind that black start services would require quarantining an amount of energy that is normally not used.

      • Craig Memery

        The network value stream poses some additional challenges as well, that go to probabilistic analysis and who has first ‘control’ over the batteries.
        Arguably, a battery used for market arbitrage is naturally likely to make some contribution to transmission on an MD day, but that value could only be captured as revenue if that portion if a contribution assigned to MD through a RIT-T… otherwise it’s like wind from a planning perspective, makes a small potential contribution but doesn’t capture the value.
        Distribution benefits may be very material at the fringe of grid too, but control and ownership (ring-fencing) issues get in the way there

  • Andy Saunders

    Might be worthwhile (seeing as you have a decent data set) to apply various charging/discharging algorithms rather than assuming being able to perfectly hit the max and min price periods every day…

    • Warwick Forster

      Andy, firstly, you’d need to look at 5 minute resolution data i.e. if the dispatch price spikes late in the half-hour, you’d miss much of the value in the half-hour. Secondly, you’d need to look at predispatch predictions during the day and then come up with a way of modelling an ideal situation with imperfect forecasts. The dataset would be huge, and the results would be subjective but worse than the example put forward here.

    • David Osmond

      Dylan McConnell has done analysis where he used AEMO’s forecast price to decide on the battery algorithm, and compared that revenue to what would be achieved with perfect foresight.

  • Peter F

    Warwick that is a very interesting analysis and if anything probably overestimates the opportunity for arbitrage because the very threat of additional competition will moderate high bids. Particularly if coupled with additional peaking capacity.
    On the other hand various people have estimated that the blackout cost SA $360m or so, maybe much more if long term interest in creating business in SA is materially affected. Some people claim that as little as 50MW and probably 100-150MW of battery power would have held the system up long enough for the gas turbines to ramp up and at least some of the windfarms to reconnect. Thus the battery pack would have paid for itself in one hit.
    How do you put an actuarial value on that

    • Warwick Forster

      Peter, I agree that the arbitrage value is lower in actuality and I did say that in the article..”A battery of 100MW would also have a material impact on spot prices in South Australia, such that it would marginally increase the lower prices and potentially reduce some of the peak prices quite significantly, thereby reducing the financial performance of the battery even further.”

      This analysis looked at arbitrage only and the market cap is above $14,000/MWh at the moment. This used to be called VOLL (Value of Lost load) to represent what households or businesses value having the power on at. Some businesses such as smelters can put a robust valuation on this in terms of lost production but for others it is much more difficult. Anyway, if the transmission or distribution system is out the price consumers are willing to pay in the affected area is irrelevant if you cannot physically deliver power.

      Indirectly, there is a value placed on outages and that is implicitly through AEMO’s reliability standard (
      i.e. it’s normally 0.002% unserved energy in a financial year. i.e. it’s far beyond a discussion here but implicitly you pay through the market and networks to achieve this reliability level.

      I don’t know about the veracity of the $360m figure given there have been a number of blackouts. It would be impossible in the current market arrangements for spot energy in SA to capture anywhere near $360m (i.e. you’d need to remove the $14,000/MWh cap to capture this in a few hours and pass it through to consumers. You also need to consider that the battery only produces at the margin i.e. 100MW share in maybe 2000-3000MW of demand in SA and you mention the contribution of gas fired generation ramping up to supply over a period of hours.

      My point is that this is an arbitration valuation only not a network support arrangement which is another valuation.

      • Peter F

        I think we are pretty much on the same page. Overall batteries will save money for SA customers but how does the battery owner capture enough of that value to justify the investment.
        It might just be that it is one of those long forgotten “public goods” that neo-liberal economics can’t assign private value to so therefore it doesn’t exist in their minds

  • George Michaelson

    Deployment should proceed on fixed-cost with arbitrage as a plus. There should be a retainer for specific functional role, and an expectation of a bid against that role. To that extent, if arbitrage alone cannot meet profit, the bid for service should make up the difference.

    Yes, people want a demonstrable profit. No, it doesn’t have to come solely from arbitrage, in a semi regulated utility function world.

  • lawrence

    A question about the 3650 cycles over 10 years; if you intend to capture 4 arbitrage events a day, wouldn’t your cycles be 4x higher?

    • Warwick Forster

      No, a daily discharge is equivalent to 100% charging and then fully discharging net of losses the capacity of the battery. So, in a half an hour you can only 1/4 charge or discharge for a .5C battery. You will find that cycles are not a useful measure, more so capacity throughput in MWh…

  • Nice work @warwickforster:disqus – thanks for bringing C-rating into this discussion.

    It would be great to hear from the Reposit guys about battery dispatch algorithms. For lower C-value installations, coming up with a good algorithm to charge and discharge batteries seems pretty tough. I’d expect that achieved value for a home battery will be *significantly* less than the “perfect foresight” model.


    Dave P.