Market power in New Zealand's wholesale electricity market: a critique of critiques

Last September, Dr Stephen Poletti published a report describing some modelling that he had undertaken of New Zealand’s wholesale electricity market.[1] He reached the startling conclusion that, in the seven years from 2010 to 2016, generators pocketed around $5 billion in 'market power' rents. This finding was announced in the midst of the government’s high-profile (and ongoing) review of the wider electricity market and, unsurprisingly, attracted a great deal of attention – as well as a significant amount of criticism.

Regrettably, the critiques of Dr Poletti’s report have invariably been superficial and frequently revealed misunderstandings of his modelling. His analysis and conclusions have consequently been widely misrepresented and distorted, which has resulted in his findings being summarily dismissed with inaccurate contentions. In this article, I consequently seek to clear up some of the confusion surrounding Dr Poletti’s report and to dispel two of the key myths that have been perpetuated about its content. 

1.          Dr Poletti’s approach and conclusions

In order to recognise the shortcomings in the various critiques of Dr Poletti's report, it is first necessary to understand how he approaches his quantitative modelling. Broadly speaking, Dr Poletti’s wholesale market analysis entails three variables: two ‘estimated’ (or ‘simulated’) and one ‘real’. The ‘real’ variable is the actual wholesale market spot prices that occurred throughout the seven-year period from 2010 to 2016. The variables that he estimates over the same period are:

• ‘simulated competitive benchmark prices’, which are calculated by assuming that all generators in the market lodge bids reflecting the variable operating costs of their plants at those times (with hydro plants’ bids representing an estimated ‘opportunity cost of water’, based on dam water level data, i.e., a ‘water value’); and
• ‘simulated agent-based prices’, which are derived via a computer model in which generators (the ‘agents’) engage in the spot market and, through a process of trial-and-error (essentially a repeated game), learn which offers will yield the highest profits and adapt their bidding accordingly, i.e., the agents seek to maximise their returns.[2] 

Estimating variable operating costs is relatively straightforward for thermal plants (i.e., it is a function of operating and fuel costs). However, as Dr Poletti notes,[3] the exercise is more complicated for hydro plants. Hydro plants do not have to incur any incremental costs to make it rain or to make snow melt. But that does not mean that water is ‘free’ or that the short-term cost of operating hydro plant is ‘zero’. That is because using water to generate electricity now extinguishes the opportunity of using that same water later. The value of that lost opportunity at any given time will depend upon several things, including:

• current storage levels (e.g., whether a storage lake is nearly full or nearly empty);
• forecast hydrological conditions which will affect future storage levels and the need to spill (e.g., whether river inflows will be high or low); and
• expected future electricity prices which will, of course, depend upon the same conditions throughout the rest of the country’s hydro schemes. 

It follows that, when each hydro plant offers its capacity to the spot market, it can be presumed to have factored into its price the opportunity cost to it of using water at that particular time vis-à-vis using it at some future point. Generally speaking, reductions in storage lake levels and inflows will increase the opportunity cost of water. In other words, the scarcer water becomes, the higher will be the corresponding ‘water values’ (i.e., when water is scarce, hydro plants will increase their wholesale bids).     

Dr Poletti calculates the ‘water values’ for his competitive benchmark scenario by building upon the approach used by Tipping et al (2004)[4] and Young et al (2012).[5] He explains that these past papers illustrated that modelling water values as a simple function of expected lake levels does a ‘surprisingly good job’.[6] He therefore employs this methodology – albeit a modified, more dynamic version[7] – to compute competitive bid prices for hydro plants. However, Dr Poletti’s report contains only a high-level description of this approach.[8]  

Any prices that exceed Dr Poletti’s simulated competitive benchmark prices are said to entail ‘market power rents’. In other words, for there to be zero market power rents in Dr Poletti’s model, the prevailing spot price must always reflect the marginal operating cost incurred by the plant that supplies the unit that ‘clears the market’, i.e., a price at which that ‘marginal generator’ covers its variable operating costs and no more (remembering that, in the case of hydro plants, this will be a ‘water value’ that reflects storage lake levels).

To be clear, that does not mean that generators never earn any profits. Rather, as Figure 1 below illustrates, in Dr Poletti’s competitive counterfactual, all ‘infra-marginal’ plants with operating costs below those of the marginal producer are paid a price that exceeds their variable costs, enabling them to make a contribution to their fixed costs.[9] Furthermore, if it is a hydro plant setting the ‘competitive’ market-clearing price, that price may also include a ‘scarcity rent’ component (a concept I return to subsequently). In each case, the rents in question do not constitute market power rents.

Figure 1: Dr Poletti’s competitive benchmark price[10]  

Market power rents arise in Dr Poletti's model only when the actual market price – or the simulated agent-based price (depending upon which is being considered) – rises above the estimated variable operating costs of the marginal producer (which, again, in the case of hydro plants, will be a ‘water value’ that reflects prevailing storage lake levels). For example, in Figure 2 below, the actual wholesale price (P*) is higher than this competitive benchmark (Pc). The difference between the revenue generators actually earned and the amount they would have earned had the marginal generator lodged a bid equal to its variable operating costs is therefore characterised as ‘market power rent’.

Figure 2: Dr Poletti’s market power rents[11]

Dr Poletti uses this modelling approach to make two key comparisons. The first is between the simulated competitive benchmark prices and the prices achieved through the repeated interactions of the agents in his simulation.[12] He finds that the prices produced by the agents seeking to maximise profits are very close to the average market prices that actually arose over the seven-year window.[13] The estimated market power rents are substantial; namely:

• the total market revenue (i.e., prices multiplied by quantities) earned by the computer-based market agents is $14.9b;
• the total market revenue that would have been earned if the generators had submitted bids equal to their estimated marginal operating costs is $9.5b; and
• the estimated market power rents under this simulated scenario are therefore $5.4b – or 36% of total market revenue.     

Dr Poletti’s second comparison is between the simulated competitive benchmark prices and the nodal prices actually observed. It again reveals substantial market power rents:[14]

• total wholesale market revenue earned by all generators over the period is $15.5b;
• the total market revenue that would have been earned if the generators had submitted bids equal to their estimated marginal operating costs is $9.5b (as above); and
• the estimated market power rents using ‘actual prices’ are therefore $6b – or 39% of total market revenue.

Dr Poletti makes two other notable findings. Firstly, he observes that market power rents have been increasing in recent years, e.g., in 2016, his calculated market power rents ($1.06b) represent 56% of total wholesale industry revenue ($1.88b).[15] Secondly, he detects significant market power rents during periods in which there is plentiful supply of low-cost base-load plant, e.g., during the 2011 ‘wet year’, when he estimates the variable operating cost of hydro to be between $0 to $10/MWh.[16] These conclusions are striking and, as I mentioned earlier, prompted strident criticisms from many quarters. 

2.          First criticism: fixed costs ignored

The most common criticism of Dr Poletti’s assessment has been that because it uses a competitive benchmark of ‘short-run marginal cost (SRMC)’ it ignores fixed costs and therefore cannot detect market power rents.[17] But that is simply not the case. This is best illustrated using a straightforward example. Suppose that to supply ‘widgets’ a producer must make an upfront investment of $100 and that, from then on, it costs $1 to manufacture each unit (e.g., for materials, labour, etc.). What is the SRMC to the producer of supplying a widget?

A very popular answer would probably be: “$1”. But that would be wrong – or, at least, incomplete. The correct answer is: “It depends”. Whenever supply is plentiful, the SRMC of producing each widget would be $1, i.e., equal to the short-run production costs – labour, materials and so on. And, if the market is competitive and the firm happens to be the ‘marginal supplier’ (i.e., the firm that supplies the units that ‘clear’ the market), then the price of widgets would also be $1 (or near to it) during those times. What is not as well-understood is that, whenever the supply of widgets became scarce (e.g., when demand surged), the SRMC would start to increase -– and so would the price. 

During those times of potential shortages, the SRMC of producing widgets would rise to whatever level was necessary to curtail demand to match supply. If there was a chance that demand for widgets would exceed the available supply, the price would increase above $1 until it reached a level at which balance (or ‘equilibrium’) was restored. Prices therefore would not need to be above the SRMC of the marginal supplier for them to deliver adequate profits to all firms – at least, not under a properly conceived notion of SRMC. Rather, during times of scarcity:

• the ‘marginal’ widget supplier could make a contribution to its fixed costs (the $100 up-front investment which, over time, would be recouped – plus a return);
• all ‘infra-marginal’ widget suppliers (i.e., firms with operating costs below $1 per unit) would make even greater contributions to their fixed costs; and
• those higher prices would also provide a potential impetus for entry and expansion, i.e., if there was perceived to be profitable opportunities on offer.

With that in mind, let us turn now to the wholesale generation market and to Dr Poletti’s modelling of it. Various parties have claimed that Dr Poletti’s analysis can be discounted in its entirety because it ‘ignored fixed costs’ and ‘nobody would invest under those circumstances’. For example, Meridian Energy (‘Meridian’) claimed in is submission in response to the Electricity Review Panel’s first discussion document that:[18]

‘…profits above short-run marginal costs are entirely expected in energy-only markets and are necessary otherwise no one would ever invest. If prices are artificially depressed so that they remain at or near short run marginal costs this will ultimately produce security of supply concerns followed by high prices.’

The error in this statement and others like it should now be apparent. Meridian seems to believe that Dr Poletti has assumed that the SRMC of generation remains constant in his model. This is the equivalent of assuming that the SRMC of producing a widget is always $1 in the earlier illustration. If Dr Poletti had made such an assumption, that would certainly be a problem. For example, if he had assumed that the SRMC of hydro generation was always, say, $10/MWh, and that this represented a ‘competitive’ price in all circumstances, then his critics would be right. But he makes no such assumption and, as such, they are not.   

It should be plain to anyone who has read and understood Dr Poletti’s report that his competitive benchmark prices are based on SRMC estimates that fluctuate depending on the level of relative scarcity. As I explained earlier, in his model, as lake levels drop and the probability of shortages grows, the ‘water values’ assumed in his competitive counterfactual prices begin to increase.[19] This is exactly the same phenomenon as our ‘widget supplier’ increasing its price to avoid potential shortages. In other words, it is precisely what one would expect to observe in a well-functioning competitive market. 

During those periods of scarcity, whenever a hydro plant is setting the market price (when it is the ‘marginal generator’) the competitive benchmark price will consequently include some ‘scarcity rents’.[20] Just as in the earlier example, those competitive benchmark prices will allow all generators to make a contribution to their fixed costs, whilst providing signals for investment in entry and expansion. It is therefore wrong to suggest that Dr Poletti’s modelling ignores fixed costs – it unambiguously does not. Rather, the analysis caters for them explicitly and the report makes this abundantly clear.    

The multi-billion-dollar sums that Dr Poletti calculates in his models therefore do not arise solely from hydro generators benignly signalling to customers when water is in short supply, since those scarcity values are captured already in his competitive benchmark prices. And, as we have just seen, hydro-plants do not need them to cover their fixed costs. Sufficient compensation should already be factored into the scarcity values enshrined in Dr Poletti’s competitive benchmarks.[21] Generators do not need even higher prices in order to earn a normal rate of return – prices at those levels are likely to deliver excess returns.

3.          Second criticism: use of spot prices

Dr Polett’s modelling considers only wholesale spot prices. However, the immediate impact of spot prices on a generator’s profits will depend also on its generation output, retail customer demand and its hedging position. For example, suppose a generator strategically re-bids its capacity and, in so doing, increases the spot price by $100/MWh. It does not follow that it will earn $100/MWh more, overall, on every unit it generates in that location over the period. To be sure, it will be paid $100/MWh per hour through the wholesale pool; but:

• if it is ‘long’ on generation[22] then, in the immediate-term, it will only earn $100/MWh more on sales not covered by its existing contracts, i.e., the uplift in price will lead to an increase in profits only on its unhedged capacity – not across its entire portfolio; and
• if it is ‘short’ on generation[23] then the near-term consequence of engineering the price increase will be that it pays $100/MWh more to purchase the additional generation it needs to meet its own commitments (i.e., generation not already contracted).

When a generator is ‘long’, the market power rents estimated by Dr Poletti will therefore overstate the immediate financial benefits from any spot price increases. And, similarly, when a generator is ‘short’, the methodology could estimate market power rents when it is, in fact, disadvantaged financially from a spot price increase. This has caused some parties to contend that an analysis of spot prices alone cannot provide insight into whether generators are earning excess returns.[24] However, these criticisms are again misplaced.

Firstly, the fact that the ‘market power rents’ Dr Poletti estimates do not necessarily represent an immediate, equivalent financial gain to the generators in question does not mean they had no incentive to increase spot prices above competitive levels over the 7-year window under examination.[25] As Dr Poletti notes in his report,[26] the major players are almost always long on generation, i.e., they are net sellers into the spot market.[27] They will therefore have frequently had a near-term financial incentive to increase spot prices to increase the returns on their unhedged capacity. 

Secondly, the longer-term incentives arising from the symbiosis between spot and contract prices must be considered. The price of hedge contracts is determined primarily by the balance of expectations as to the level and volatility of future spot prices.[28] Ergo, if average spot prices are seen to be increasing – e.g., because of the short-term incentives described above – this can usually be expected to result in higher contract prices.[29] Specifically, if generators can influence spot prices, this will be normally reflected in the contract prices that they are prepared to accept. Counter-parties will then have to choose between:

• remaining unhedged, and being forced to pay the high spot prices that may result from generators acting on the financial incentives described above; or
• entering hedge contracts and paying prices in which those high spot prices they would otherwise be forced to pay are ‘averaged’ over the lives of the contracts.

In other words, there are two compelling reasons why a generator with the ability to do so may seek to increase the wholesale market spot price; namely: 

• to increase the price received for its unhedged capacity (i.e., if it is long in generation), thereby increasing its near-term returns; and
• to affect the balance of expectations regarding future spot prices, thereby potentially increasing future contract prices for its hedged capacity.[30] 

I consequently do not agree that a diagnosis of market power requires contract prices and positions to be taken into account. To be sure, this additional information may be helpful, but it is not necessary.[31] In my opinion, an examination of spot prices over time can, in and of itself, provide a strong indication of whether generators are offering their capacity at competitive levels and whether market power problems potentially exist. Dr Poletti’s decision to focus solely on spot prices therefore does not compromise his analysis.  

4.          Conclusion

Once Dr Poletti’s modelling is comprehended properly, it is no longer possible to summarily dismiss his findings with casual throw-away lines like: “but what about fixed costs and scarcity rents?” Furthermore, it becomes rather more difficult to conceive of a benign explanation for the $5.4 to $6b sums he estimates over his sample period, especially considering that:

• his analysis accounts explicitly for the scarcity value of water and that, in our hydro-dominated system, scarcity is almost always synonymous with low lake levels;
• his estimated market power rents represent around 35-40% of total wholesale revenue, on average, over the seven-year period he examines (a striking proportion); and
• the rents he calculates have arisen even at times when there is no apparent scarcity, e.g., during the 2011 ‘wet year’ (after all, scarcity rents require scarcity).

Of course, that is not to say that there might not be other problems with the modelling that could come to light following a more thorough examination. However, as it stands now, Dr Poletti’s results appear to raise significant questions about the degree of competition in the generation market that have not been fully addressed. 

[1]    Poletti, S., (2018). Market power in the New Zealand wholesale market 2010-2016, University of Auckland (see: here) (hereafter: ‘Poletti (2018)’).

[2]    Specifically, the computer agents are firms who own generation assets. Each period the firms offer all their available capacity into the spot market. The firms will typically choose to offer different units at different prices. Some of the larger generation units are allowed to offer up to four tranches of prices/quantity bids by splitting them into units with smaller capacity. The offer prices are found by trial-and-error through a learning reinforcement algorithm. Each period, the firms draw offer prices for each of their generation units from a probability distribution which is updated at the end of the period using reinforcement pay-offs. The market is cleared, and profits computed. Actions that return high profits have an increased probability of being played the next round, with the process repeated 1,200 times to simulate prices for a single 30-minute trading period. By the end of the simulation, the computer has ‘learnt’ which price offers are likely to yield the best profits given the other firms’ expected actions and the simulation ends. See: Poletti (2018), p.10.

[3]    Poletti (2018), p.7.

[4]    Tipping et al (2004), The incorporation of hydro storage into a spot price model for the New Zealand electricity market, Presented at the Sixth European Energy Conference: Modelling in Energy Economics and Policy. Zurich.

[5]    Young et al (2014), ‘Can Agent-Based Models Forecast Spot Prices in Electricity Markets? Evidence from the New Zealand Electricity Market’, Energy Economics, 45, pp.419 – 434.

[6]    Poletti (2018), p.8.

[7]    The water value curve is computed as a function of the actual lake level, compared to the mean, for any given day. Dr Poletti states that his methodology seeks to ensure that the lake levels over the course of the year are ‘dynamically consistent’. If lake levels start to get low in the competitive benchmark, the higher water levels lead to less dispatch which acts to correct the tendency seen in previous studies for the competitive simulations to dispatch ‘too much’ hydro. See: Poletti (2018), p.9.

[8]    For example, Dr Poletti does not provide a detailed description of the derivation of his water value curves. Rather, the description contained in footnote 8 above largely represents the extent of his explanation. Given the importance of this aspect of the modelling to Dr Poletti’s findings, the report could have benefitted from a more comprehensive account of this element.

[9]    For example, if the marginal generation plant in a pricing period is, say, a gas peaker with high start-up and fuel costs, then base-load plant with lower marginal operating costs (e.g., a large hydro plant with plenty of stored water) will earn positive economic profits during that timeframe. 

[10]   Although Figure 1 depicts a downward sloping demand curve, Dr Poletti’s modelling assumes implicitly a ‘perfectly inelastic’ (i.e., vertical) demand curve.

[11]   Ibid.

[12]   Poletti (2018), p.39.

[13]   Over the 7-year period, the average simulated price is just $5.1/MWh below the actual average spot price.

[14]   Poletti (2018), p.41.

[15]   Poletti (2018), p.41.

[16]   Poletti (2018), pp.37-38.

[17]   See for example: Options paper, p.21.

[18]   Meridian, Electricity Price Review: Meridian and Powershop Submission, 23 October 2018, p.52.

[19]   This is because, as water becomes scarcer, the opportunity cost of using it to generate electricity at that point in time increases. See: Poletti (2018), p.9.

[20]   Indeed, in New Zealand’s hydro-dominated system, low lake levels are typically synonymous with scarcity. As lake levels start to drop, the probability of shortages starts to rise – almost by definition.

[21]   Dr Poletti’s competitive benchmark prices for non-hydro plants (e.g., thermal plants) do not factor in a ‘scarcity’ element in quite the same way. Instead, the operating and fuel costs are assumed to be constant throughout each year, and these plants’ bids are assumed to always reflect those short-term variable costs. However, even if scarcity can, in principle, arise also from thermal plant fuel shortages and transmission/plant outages, these considerations are far less significant, in practice, than the scarcity value of water, which Dr Poletti’s model captures. Furthermore, the way that the analysis has been undertaken means that this potential limitation affects only one of his comparisons. Specifically, neither Dr Poletti’s ‘simulated competitive benchmark prices’ nor his ‘simulated agent-based prices’ are influenced by scarcity arising from factors besides low lake levels, e.g., they are unaffected by transmission and/or thermal plant outages. There would therefore be no conflation of scarcity and market power rents in the assessment.

[22]   A generator is ‘long’ if its wholesale revenue from generaton and derivatives is greater than its wholesale costs from purchases and derivatives, i.e., if it is a net seller of generation.

[23]   A generator is ‘short’ if its wholesale revenue from generaton and derivatives is greater than its wholesale costs from purchases and derivatives, i.e., if it is a net seller of generation.

[24]   See for example: Sapere Research Group, Electricity Sector Review 2018, Prepared for Business NZ Energy Council, p.46.

[25]   I note also that at no stage does Dr Poletti claim that this represents the incremental, overall financial return that generators would have earned in those periods

[26]   Poletti (2018), p.11.

[27]   If a generator with market power hedges all its capacity, it exposes itself to substantial spot market risk if it cannot deliver that capacity. For example, if one of its generating units experiences an unplanned outage, this may result in a material increase in the spot price during that period. If it is has entered into contracts for the exchange of funds by reference to its unavailable capacity at, say, $60/MWh, and the spot price during that trading interval is $1,000/MWh, then it must effectively procure that capacity at the prevailing spot price ($1,000/MWh) and sell it at the contract price ($60/MWh). It is the potential for such losses that induces generators to be cautious about becoming fully hedged.  

[28]   If this were not the case – and the price of hedges was out of line with expectations of future market prices – then profitable arbitrage opportunities would arise to close the gap.

[29]   As I explain in more detail in a recent report, this relationship may not necessarily be perfect, but it is reasonable to anticipate a strong symbiosis. See: Green (2018), Economic review of electricity generation and retail market issues, A report for Vector, October 2018, p.4.

[30]   A generator’s incentive could also be to maintain contract prices at existing levels if, for example, it had already been affected by such conduct, i.e., in previous periods.

[31]   An analogous conclusion was accepted by the Australian Energy Market Commission when it was exploring the issue of wholesale market power in Australia’s National Electricity Market. See: Green et al, Potential Generator Market Power in the NEM, A Report for the AEMC, 22 June 2011 (available: here).