INTERNATIONAL ACADEMIC SYMPOSIUM
ENERGY SUSTAINABILITY AND COMPETITIVE MARKETS Parc Científic de Barcelona. Auditorium Barcelona, January 29, 2013
Organising institutions:
CHAIR OF ENERGY SUSTAINABILITY AT THE UB
Sponsor:
In this talk1 I plan to offer you an overview on competition in energy markets, paying close attention to the role of long-term contracts or what we sometimes call ‘forward contracts’. Before explaining exactly what I mean by that, I’ll review the literature and then discuss the work I’m currently conducting in this area, as well as some open questions for further research. Motivation The main question, which is also very much a policy question, is this: Do long-term contracts or forward contracts enhance competition in markets? And what about electricity markets in particular, where most of the attention has been focused? A basic definition of what I mean by a ‘forward contract’ is the idea that part of your production is contracted in advance and so when you come to the spot market the basic mechanism at play is that you have incentives to be more aggressive. The question, however, which is also part of what we are trying to address here, is whether in practice this is actually so. Of course there are many reasons for forward contracting, including hedging reasons coming from supply/demand uncertainty, but my focus here shall be exclusively on the strategic motivation, meaning that you want to gain a better position in the spot market (in the end, of course, you may come out of this worse off, and we’ll also be considering that possibility). Finally, the talk will generally examine models that assume complete information and no uncertainty, although I’ll briefly comment on what happens if you relax some of those assumptions. Further motivation and organization of the talk When we talk about the electricity market I think that the basic assumption is that we’re talking about a homogenous commodity and few players. In a moment I’ll be looking at Allaz and Vila’s seminal work of 1993, where they first demonstrated why forward contracting can be procompetitive; for the moment, though, I’m basically going to describe the shape of the rest of the talk, explain where this pro-competitive result comes from, what produces that result and what the evidence tells us. I should also like to wonder why in the electricity market here in Spain, in the UK, the US and other countries around the world forward contracting is an important factor in market restructuring because of its potential to produce better results in terms of competition, and that is also why I want to discuss what happens when we relax some of the assumptions of the basic Allaz and Vila model. For example, what happens if you introduce repeated interaction? Because in electricity markets we know that the competition between firms is not static but repeated and that firms are engaged in a game of many periods, as it were. So here are some open research questions that we can briefly consider. First, there’s the notion that in some cases and in some countries firms have been forced to forward contract but then there’s also the question of when they have endogenous incentives to do this and what the empirical evidence tells us about that. The question remains open, I’d argue, because in most markets you’d expect to see companies that had been forced to take this forward position and yet we cannot confirm that this is what they’re actually doing; and the little research we have has not yet provided clear answers about this important topic, either. Second, how intensively do firms compete for these long-term contracts and forwards? Almost all the literature discusses whether the competition in the spot market is more or less competitive 1
See the slide presentation at www.ieb.ub.edu/files/Slides%20-%20Montero.pdf]
1
as a result of these forward contracts but we know little about how competitive these long-term contract markets are (incidentally, work on data from Chile that I’ve been engaged in would suggest that the news is not that good). A building block: the Allaz and Vila pro-competitive result (JET, 1993) [www.ieb.ub.edu/files/Slides%20-%20Montero.pdf, Slides 5–9] The basic Allaz and Vila model is a duopoly Cournot model preceded by a forward market, which means two periods (t = 0,1), two firms i and j and constant marginal costs c (we’re not going to introduce capacity constraints yet, though you may want to). At t = 0 firms are allowed to sell these forward contracts and can take either a short position, meaning they’re selling quantities in advance, or a long position, meaning they’re willing to buy these quantities. You can leave that open and the part of the equilibrium is that they want to go short in the market. And after the forward positions are sold or taken, you open up the spot market and the firms come to the spot market with these forward positions already taken and we denote the production by qi and qj. We’re assuming that the demand in the spot market is linear, although this is for convenience’s sake and simply allows us to have closed-form solutions for the equilibrium. What is the equilibrium in this game, in which we have these two stages? We use a concept of subgame perfection here so let's go to t = 1. Given these forward positions, F1 and F2, these are the profits or pay-offs for the firms and this is where you have to understand what a forward contract is. When you come to the spot market your total production is qi and this is going to be the spot price; but part of your production, fi, was already sold in advance and so your profits in the spot market are only this part, this residual production. If you’re fully contracted, there’s no extra profit in the market because you already contracted everything. And when that happens, we know in a Cournot model that the reason you aren’t fully aggressive or don’t implement the perfectly competitive solution is that you’re concerned with the inframarginal units. Here there are fewer inframarginal units and so you end up being more aggressive in the forward market. What basically happens is that your marginal revenue goes up. And what’s the result of the equilibrium in the second stage? Taking these forward positions as given and assuming that fi and fj are practically the same, you produce more than you would under the typical Cournot solution and the price is lower (remember that Cournot for this linear model is simply a minus c third). This is precisely the pro-competitive result of forward contracting that Allaz and Vila and others discuss: the fact that if firms contract in advance they’re going to be more aggressive in the spot market. In many cases this forward position has been imposed by the regulators exogenously and this has happened in Spain, in the UK and elsewhere, even. As part of the restructuring of these markets, these suppliers or generators have vertical arrangements with the retailers which basically work as forward contracts, because the price at which you’re going to sell part of your production is already fixed. So it works in the same way and produces the same incentives as in the spot market. If this is the pro-competitive result, now we’ll move to what is perhaps a more interesting question. If you’re going to be more competitive in the spot market if you sign these forwards in advance, perhaps there's no reason for you to do it. So what happens in the equilibrium? Now we go to the subgame at t = 0 and as a function fi and fj this is the equilibrium in the spot market, given those two quantities, which we derived before. This is the equilibrium quantity of firm i given those forward positions, and the same here, and this is the forward price. So you can divide this profit function or payoff at t = 0 in two: what I’ll call the Cournot profit, on the one hand, and the arbitrage profit, on the other. These are the extra profits you can make if at t = 0 you signed a very profitable or highly-priced forward contract and then in the spot market the prices 2
were much lower. Why? Because you simply go to the spot market to cover your position. If this is higher, then you lose money. This, in effect, is the arbitrage profit. But what happens in this model where you have rational expectations, where you are the generator taking a short position and there is therefore someone else who must be taking the long position, be this a consumer or an arbitrator? We’re assuming that you have rational expectations so you anticipate the impact those positions will have on the spot market and in equilibrium those prices have to be the same. All that’s happening here is that by signing forward contracts you’re changing your Cournot pay-off, as it were. But we already know that if you sign forward contracts you’re going to be worse off, so why would you actually choose to do this? Suppose that one firm hadn't signed any forward contract. If you as a generator or supplier had the opportunity to sign one, it would in fact make sense to sign. Why? Because this would give you a bigger share of the oligopoly rents in the spot markets. Remember the Stackelberg model in which there are two Cournot players and in which you allow one to move first: in this case, you want to be that one, in order to get a bigger share of the oligopoly rents. Using the forward contract is a way to do that and in fact that's exactly why you want to go forward: you want to get that ‘Stackelberg force’ and it’s because of that force that you want to go forward to get a bigger share of this oligopoly rent in the spot market. But the problem is that both your firm and the other firm want to go forward so that in the equilibrium you have a prisoners’ dilemma where both parties end up signing forwards and both are worse off. Here you have a pro-competitive result that is reinforced by the fact that there are endogenous reasons for firms to go forward. As we know, if they go then it's more competitive but it’s not only that; you also have endogenous reasons for going forward. So far we've only discussed two firms and one contract period or round of contracting. But what happens if you relax those two assumptions and introduce more firms? If you introduce a large number of firms, is this Stackelberg force going to be cancelled or nullified because, as the market becomes increasingly competitive, fewer oligopoly rents remain on the table and here are fewer incentives to go for them? Well, the answer is actually the contrary. As you introduce an increasing number of firms, even if those oligopoly rents become smaller there are still more incentives for you to sign forwards to get control of a portion of those rents. So you increase those incentives and when you get to the limit, where you have a very large number of firms, they all end up fully contracted and you implement the fully competitive solution. That, I’d argue, is an important prediction of this model: the more firms you have, the more contracted you will be. It may not be in Allaz and Villa but as an extension it’s simple enough for students to do (as I confirmed in an examination with my students last year). This next part is in the model in Extension II: What happens if you start adding more rounds or periods and firms don't only have one opportunity to sign these contracts before the opening of the spot market? When this happens, and in contrast to the previous model, you don't go fully competitive. Instead, you go halfway between the Cournot solution and the price equals marginal cost. There are still oligopoly rents on the table, however, so what if now, before that game of two stages, you open a new opportunity to sign forward contracts? Now you have a Stackelberg incentive to go for those lower rents but still some rents on the table. The question is: Do you use that incentive? Of course, and you’ll use it now for the same reason you used it the first time, and that’s this new result: that whenever you have an opportunity to sign forward, that’s what you’ll do.
3
And if you start adding an increasing number of periods, you become fully contracted. So N is the number of periods or contract rounds before this bottom market is opened, and F is your cumulative forward position. You can cancel the positions as you move on or sell short first and then sell long or do any number of things before the spot market is opened; but in equilibrium you’re going to be gradually signing an increasing number of forwards and that’s the proof of the paper. And what’s important to understand is that as you go to this price equilibrium function, you see that as you add more periods this goes to zero, so you go to marginal cost. The Allaz and Vila predictions The first two predictions are very important, I’d argue, because they reflect what’s natural: because when companies are thinking about next year’s spot market opening there are many opportunities for them to sign these forwards. So based on Allaz and Villa you’d expect certain things and this might be a good moment to summarise the predictions. Prediction 1 is that whether you’re contracted for endogenous or exogenous reasons, larger the contact coverage (short positions), the more competitive the spot markets will be. But my feeling is that the related Predictions 2 and 3 are more interesting: firms have possible a large number of endogenous incentives to sell forwards and the amount of contracting is increasing in the number of firms, so you’d expect them to be fully contracted at the opening of the spot markets. What the evidence tells us The empirical evidence provides us with mixed results because not all three Allaz and Vila predictions are well corroborated. If we look at the restructuring of the UK power market in the early 1990s, we’ll see that in her discussion of market competition and the role of forward contracting Wolfram (AER, 1998) found little evidence of Prediction 1. According to Wolfram, this was because by 1993-94 there was a reduction in contract coverage and the prices didn’t change that much. So contract shouldn’t have an effect, because if you have fewer contracts, she observed, you expect prices to go up. Still, something was happening because the prices were much lower than the Cournot prediction, and Wolfram attributed this to regulatory vigilance, threats to divest companies that wouldn’t behave or even the threat of entry, meaning that companies were pricing at a level at which entry wasn’t profitable. In theory it’s like a limit price, although I can’t really support this (for limit pricing to be effective you need a lot of asymmetric information and I’m not sure there was enough of that). The regulatory threat was real, certainly, but the role of contract coverage was also important, as Green (JIE, 1999) subsequently observed. My problem with this is that a portion of these contracts were exogenously imposed on firms. To return to the prediction by Allaz and Vila what we need to understand is that it’s in the interests of the firm or supplier to go out there and sign these long-term contracts. In many cases you have buyers coming to you to sign contracts. The Allaz and Vila prediction holds that it’s the companies and generators that go out there to sign contracts as a commitment device to gain a better position in the spot market; and that very important part of the prediction is not one we always keep in mind when we talk about forward contracts: the Allaz and Vila model understands that the forward market is supply-driven, not demand-driven. And yet the truth of the matter in most cases, as far as I can see, is exactly the opposite: it’s a demand-driven market, with large buyers coming to these power generators looking for contracts. But as I said, companies in the UK were highly contracted because of these exogenously imposed conditions. To some extent Green disagrees with Wolfram, arguing that prices were 4
lower than the Cournot prediction because of contracting, but then his objection is offered more to serve the purposes of debate rather than as the hard-and-fast results of econometric testing. And in a more recent paper, Sweeting (IEJ, 2007) observes that prices for the power pool were actually higher than in the Cournot prediction and that there was evidence of collusion in the UK market in more recent data, despite a coverage of over 90%. So that’s bad news. For the Australian market, Wolak (IEJ, 2000; JAE, 2007) observes results that came very close to Allaz and Vila’s Prediction 1: namely, that firms were very competitive in the market because of this large contract coverage. But Predictions 2 and 3 were not tested there and we need to remember that there are endogenous reasons for firms to go forward. It was only Prediction 1, and once again it’s because these firms were basically contracted for exogenous reasons. So we return to that question of whether what is driving contracting is a supply-side force or a demand-side force. I’d argue that it’s mostly a demand-side force. In a very important paper, Fabra and Toro (IJIO, 2005) found that in the Spanish case the Allaz and Vila prediction wasn’t operating and that, on the contrary, there was evidence of market collusion. The situation may have shifted since that data was collated, in the late nineties or early 2000s, but the important thing is that contracts were being exogenously imposed on firms, although they could contract more, because they were not fully contracted, and they didn’t do that. For the American market, three markets were reviewed by Bushnell, Mansur and Saravia (AER, 2008): the California market, the New England market and the PJM market (Pennsylvania, New Jersey, Maryland and DC). They found that when you corrected the Cournot prediction by this contract coverage, in many cases instead of having forward contracts the contracts had this vertical arrangement because some suppliers already had arrangements with the retailers and were basically selling part of their production ahead, at a fixed price, which in turn operated exactly as a forward contract would in terms of incentives to bid in the spot markets. So Cournot behaviour corrected by contract coverage was found to be a very good predictor, as the figures in Slides 17–21 from Bushnell et al. demonstrate [www.ieb.ub.edu/files/Slides%20%20Montero.pdf, slides 17–21]. In the California figure, the dash line is a perfectly competitive solution, the black solid line is the actual prices and the grey line is the Cournot prediction. They come very close, and here there are no contracts. But here you have contracts, for example. This grey line is the Cournot prediction, these are the actual prices in New England, and the dash line is the perfectly competitive solution. So what happens when you correct the Cournot solution for this contract is that firms become more aggressive in the spot market and this is exactly what you get. So you’re here and you’re correcting, and this grey line is now Cournot corrected by contracts, very close to what the data is telling you. But there’s still an open question for Bushnell et al., which is why didn’t firms sell more forward contracts? Because only 70% to 80% were fully contracted, but you still have some oligopoly rents. Why didn’t they contract more? This is Allaz and Vila’s Prediction 2 and 3. There’s actually no answer in the paper and what you see here is all they do. So for me that’s still an open question. They didn’t go for more; they just stayed there. We then have van Eijkel and Moraga-González (2012), the only paper I’ve seen — but correct me if you’ve seen another — that specifically tests Predictions 2 and 3 (i.e., these endogenous reasons for selling forward contracts). These authors examine the Dutch wholesale market for natural gas, and they test using the fact that the number of players in the market changed from year to year. Now if you change the players from year to year and you remember Allaz and Vila’s Prediction 3 you may expect the first to be contracted more. Unfortunately, however, that prediction is very weak because you have many rounds of contracting. The Allaz and Vila 5
predictions are very simple: you should be fully contracted. So in an important sense the variation we’re using – that more wholesalers in the market leads to more contracting – is not a very powerful one because you don’t actually need that many if, according to the prediction of the model, you can have just two or three and still be fully contracted. This is further compounded by the opacity of the market, which is an issue we’ll return to. Because if the forward positions are not entirely observable, then this disregards the original Allaz and Vila model’s basic assumption that you observe your rival’s position and that’s why you can use this as a commitment device. If you fail to do that, you lose this credible device. And if positions aren’t fully observed, you’ve completely lost that. If they’re partially observed, you still have something. So this paper also needed to deal with how transparent the forward market was and tried to answer that as well. Now let’s turn to experiments. In this area Le Coq and Orzen (JEBO, 2006) support Allaz and Vila’s Prediction 1, albeit only to a point, and Ferreira et al. (2010) found that even if you have a repeated interaction, players tend to play the static game. In a more recent paper, however, Schubert (2012) observes that when you allow firms to be engaged in this interaction they use the forward contract to sustain collusion. How can we summarise the literature? Some markets support Prediction 1, which is that if you’re contracted you’re going to be more aggressive; but this is more common in the American markets than in the UK (either at the beginning or more recently) or in Spain. And while Predictions 2 and 3 haven’t been tested directly, I’d argue that the evidence points to these predictions not working: when firms have been contracted they don’t go for more; they go to the point at which they are being exogenously imposed upon by regulators or demand-driven. Relaxing some of Allaz and Vila’s assumptions I’d now like to relax some of the assumptions in the basic model to see whether we can remain consistent with these empirical findings. This can be done in different ways: adding demand uncertainty, making the forward position not fully observable, making firms compete in prices instead of in quantities, allowing for repeated interaction, etc. Demand uncertainty If you add demand uncertainty the model hardly changes at all and if you assume risk neutrality producers contract as before. No observability of forward positions As Hughes and Kao have shown (IJIO, 1997) , if you make the positions unobservable you simply lose the Allaz and Vila results because this instrument no longer works as a commitment device. And this finding is also in van Eijkel and Moraga-González (2012). When you boil down the technicalities, the reason is basically this: you’re right to make certain conjectures about why you’re playing, but when it comes to the spot market you don’t know your rival’s forward position; and if you know your rivals are going to play Cournot in the spot market, you gain nothing at all by signing forward contracts. You sign forward contracts to show them you have some forward position; and if they know you’re going to be more aggressive than they are, then your rivals are going to produce less. This is the substitution of quantities. But if you can’t convince a rival to produce less, you’re much better off going back to Cournot. You don’t need
6
the positions to be fully observable, but remember that some of the Allaz and Vila positions are somehow observable. Prices v quantities v both What happens if you relax the Cournot assumption in the spot market? Mahenc and Salanié (JET, 2004) show, inevitably perhaps, that if you relax the assumption of quantity competition and replace this with price competition, your spot variables change from being substitutes to being complements. That is, if one firm lowers its product price, the other firm’s bare response is to lower its own and when the first firm has substitutability and increases its production, the other firm’s bare response is to reduce its production. That is at the heart of this model. If you switch the mode of competition from quantity to prices, the result is exactly the opposite. You still have Prediction 1 (that if for whatever reasons firms are fully contracted they are going to be more aggressive); but now when we come to the first stage, when you have to decide how many contracts to sign, you want to be soft. You actually want to take long positions, not short ones, and that is because the spot market competition has changed. This is Fudenberg and Tirole’s ’puppy-dog ploy’ and ‘top dog’ results (Fudenberg and Tirole, 1984), where you decide to under-invest and go long instead of short, making it even worse for the competition in the market. Mahenc and Salanié also observe evidence of market collusion in such quarters as the Pancafe coffee cartel, where suppliers cornered the market by taking long instead of short positions. But that’s a dynamic story while this paper is a static story and is simply changing the mode of competition in the spot market. Green (JIE, 1999) goes halfway between price v quantity and uses what is basically a schedule in which firms bid these supply functions in the spot market: i.e., show willingness to buy or sell this quantity at this price. And if the schedules are linear, then your schedule in equilibrium is going to be independent of the forward position of your rival. In the Allaz and Vila model, by contrast, if your rival already has taken some forward positions then you want to produce less. And that’s your interaction and the reason why at the first stage you want to sign forwards contracts. In this model, if you’re submitting the supply functions to the market in equilibrium, forwards do nothing because they don’t affect how your rivals behave in the spot market. This is a very strong assumption but it’s only because of the linearity imposed in the equilibrium. Remember, in this supply function equilibrium you have an infinite amount of multiplicity and the Cournot is the upper bound, the worst you can get in terms of competition. But these models could be pointing to the fact that firms don’t want to go fully forward because the way they compete in the spot market is not Cournot, but something quite different to Cournot. And this could explain what was happening when we reviewed the evidence before. Repeated interaction and collusive behaviour In Liski and Montero (JET, 2006) we return to the notion that forward contracting may be helping firms to sustain collusion, and this would be consistent with the description of Spain in Fabra and Toro (IJIO, 2005) or with the story in the UK. The question is this: What happens if firms can sign forwards in every period, engaging in what is basically a repeated interaction game? Does this allow them to sustain collusion if they lower the critical discount factor over which they can sustain monopoly profits? If it does, you might conclude that contracting allows collusion or makes collusion easier, if you decrease the critical discount factor at which you can sustain collusion.
7
The results don’t depend on whether the form of competition is in prices or in quantities, and two effects interact in this repeated interaction game: the pro-competitive effect and the pro-collusive effect. The first can be explained as follows. If all your production has been contracted ahead, one reason why you don’t want to deviate from a cartel agreement is that you have collusive profits in the future that you want to protect; but if all those collusive profits or monopolies have already been signed ahead, that profit is in your pocket and you have more incentives to deviate. That, in essence, is the pro-competitive effect and if you forced firms to contract a lot, this effect would still be present in this repeated interaction game. But then there are also two pro-collusive effects of forward contracting. The first comes into play when an agreement is weak or collusion is difficult to sustain because there are too many incentives to deviate, and one way of reducing the number of incentives is to use forward contracts. At the moment of deviation, consider a typical Bertrand game: if you deviate, you undercut, by epsilon, and you get the full market. But if part of the market is already contracted, you obtain less. So you can reduce the incentives to deviate by contracting some of that demand ahead and this, in a nutshell, is the first pro-collusive effect. The second is that the punishment phase is now harsher. Why? Because we know from Allaz and Vila that if you enter the punishment phase it’s going to be more competitive and that’ll help you sustain collusion. So we have these two interacting effects: on the one hand, if you contract too much you’ll experience the pro-competitive effect, but then you’ll also have the pro-collusive effect. And while we’re discussing this paper, if contracting is endogenous then firms will choose an optimal number of contracts that allows them to sustain collusion in cases where this would not have been possible under simple spot competition. One final observation, however: if firms are exogenously required to contract a lot, for whatever reasons, then the pro-competitive effect is going to dominate anyway. Contracting for an exhaustible-resource (spot) market In Liski and Montero (JET, 2006) we also considered exhaustible resources. Remember that with the Allaz and Vila mechanism, you finally have more production in the market. So in some cases you have commodity markets for exhaustible resources in which you have a fixed supply and here, you might argue, the Allaz and Vila mechanism can’t be operating because you can’t bring more output to the market (of course there are different periods in which you deliver this market or you allocate this fixed supply, but the total amount is fixed). Actually, the mechanism is operating here; it’s just operating differently, as we shall now see. Slide 36 [www.ieb.ub.edu/files/Slides%20-%20Montero.pdf, Slide 36], describes the equilibrium price path for an exhaustible-resource (spot) market and indicates the time, price, and backstop technology price at which you switch to the substitute. Here you have the perfect competition solution, which means prices go up by the rate of interest in an exhaustible resource (assuming no extraction cost, to simplify things). This is very useful, for example, if you want to understand how a market for pollution permits would work, because you would store these permits and at some point you would hit the long term. (Indeed, we started this research by trying to understand exactly that: how a large firm in a permanent market for storable pollution would exercise market power.) Finally, in the oligopoly solution it’s not the prices that increase at the rate of interest but the marginal revenues (prices increase at a somewhat lower rate). This is the price path in the oligopoly solution, and the exhaustion comes even later on. How can you use the forward contract to improve your position here? If you look at these two price paths, along the oligopoly path the most profitable periods are the early ones. These prices 8
are in present values because prices are going up but are somewhat lower than the interest rate, and the present value of these prices is down here. Let’s say that you want to convince your rival you’re going to be producing more in the early periods than in the later ones, and that’s where you can use forward contracts. The forward contract allows you to move production credibly in equilibrium from the future to the present. But both firms do that and, eventually, they both become more competitive. Therefore, and as I observe above, the Allaz and Vila mechanism is still there; it’s just operating differently. Small but uncertain cost asymmetries This comes from my most recent research, conducted with a number of considerations in mind. First, all the empirical evidence shows that firms don’t have incentives to forward contract for endogenous reasons. In other words, there’s no supply-driven forward contracting here. There are only two possible reasons why firms are contracting forward: either the regulators have been imposing this behaviour on them or it’s demand-driven by a large industrial buyer that they need to contract in advance. But my view of this is that the Allaz and Vila mechanism is not operating. You might argue that this is because firms are colluding and don’t need to contract much, or because they are simply not playing Cournot, and those might be reasonable arguments. But let’s suppose that something else is happening. Let’s suppose that firms are playing Cournot. Let’s return to the arena of Allaz and Vila, the most favourable arena because that’s where firms contract endogenously and fully. And let’s shake up that model in a simple but realistic manner by adding a dash of uncertainty. First, remember you’re signing these forward contracts at one point in time but then you’re learning your types a whole year later, right before production, so you’re not sure about your marginal cost. Let’s suppose your marginal cost when you’re signing these forward contracts could be either higher or lower, and you have these two firms if you want, but you don’t know. Or you can also assume some private information about your actual cost (I haven’t actually done this but I believe it works). Now let me go back to the uncertainty case and assume that c upper bound and c lower bound are very close. We’ll introduce a little uncertainty, but not much. You may think this is basically taking us back to Allaz and Vila, but it isn’t. The probability that your cost is low and the probability that your cost is high are both described in the figure, so these are just two possible events. And what happened to the Allaz and Vila model? They reverted. Why? First of all, when you introduce this cost differential, you may be limiting that Stackelberg incentive you had for signing forward contracts, which was that you wanted a bigger share. When you introduce a cost differential, at some point that ‘Stackelberg force’ disappears. And again, why? Because if you have much lower costs than the other firm there’s nothing you need to worry about when the other firm comes to the market. As you start to contract more and as the market itself becomes more competitive, the initially very small cost differential becomes very large. And at some point you decide that you don’t want to contract any more if you expect that you could be either below or above your rival’s cost, and that is what kills the Stackelberg force. Slide 38 [www.ieb.ub.edu/files/Slides%20-%20Montero.pdf, slides 28] describes my prediction. Remember, the Allaz and Vila prediction was that if you let the number of firms M go to infinity, this is how much you are contracting and this is how much you are producing. This was one firm, so when you increase the number of firms the Stackelberg force becomes very much stronger and you want to fully contract. In equilibrium, that is what would happen. If you add this tiny dash of uncertainty (if this delta is the difference between these two costs) and you let that uncertainty go almost but not exactly equal to 0, you completely kill the Allaz and Vila force. 9
With just a small perturbation of the game, you totally revert it. And that is because the game is based on this Stackelberg force; but if you introduce tiny asymmetries in cost, you completely eliminate that. And this might give us good reasons to argue that Allaz and Vila’s Predictions 2 and 3 don’t actually work. Or that they work in their model form but not in reality and so the only reason we see heavy contracting is for exogenous reasons. Competition for long-term contracts: auctions v bilateral trading Now let’s turn to my research on the competition for long-term contracts, which I’ve been conducting with data from Chile. In Chile, instead of a power pool for the spot market, we have a merit order system in which the regulator decides which units have to be dispatched. But we have a decentralized market for long-term contracts, in which large buyers come to their suppliers for contracts. And the main question in my research is this: How competitive is that market? What’s interesting about the Chilean experiment is that while retailers used to have to negotiate their energy bilaterally with the suppliers, in 2006 the authorities started to enforce auctions. Each retailer has to open an auction to cover all its demand needs and all the generators come to these auctions to bid. So in a sense, in recent history Chilean generators have been engaged in two market institutions: from before 2006 to the present time they’ve been engaged in bilateral trading with large buyers, but since 2006 they’ve also been bidders in auctions. Do the generators behave differently depending on which market they’re in? In terms of opportunity cost, they don’t: the opportunity cost for these firms is to go to the spot market to buy energy from another supplier to cover their contracts, so the opportunity cost is the same and just the market institution is different: in one, there’s bilateral trading and the negotiation is with each large buyer separately, and in the other all the generators go to retailer procurement auctions. But if the question is whether they compete differently, then our research suggests they do and that even though these results are still very preliminary, that difference is substantial. The figures in Slides 42 and 43 [www.ieb.ub.edu/files/Slides%20-%20Montero.pdf, slides 42– 43] come from information that was made available to me from the national energy commission on which I sit. Under normal circumstances such data would be deemed confidential but I was allowed to go ahead and use this particular data for my research and I’m happy to be able to present it to you here. In Slide 42 the blue dots are the contracts between the suppliers and large buyers and these are the contract prices used to set the prices for the residential sector. Because under the premise that here you have good competition, the residential sector should get the same prices. The red line is the marginal cost in the spot market. This is the average (every six months) at which this dispatch market clears every hour. You would expect the two to be very close especially during the first period of time, which for Chile was a very calm period for natural gas, and yet one is much higher. So why are these contract prices not closer to these others? As we move on in time, here we have the Argentine natural gas crisis of 2004 when they basically turned off the tap and we had to start using diesel power, and that’s why the marginal cost goes up so high. And finally, we are now somewhere back here. The question, however, is whether these contract prices are close to the opportunity costs, which is what you actually expect for this marginal cost price. Now here you have the same data for the retail auctions. Slide 42 describes the prices and Slide 43 describes the comparison between the 10
two. The blue lines are the prices of the contracts being signed in bilateral negotiations between the suppliers and the large buyers, and the red dots are the results of the auctions, in which you have all the generators coming to auctions for the retail companies. And the two figures show you what a big difference there is. One question often asked here is why there’s so much price dispersion and the reason is that some of these contracts are indexed to the marginal cost of the bar at which the energy has to be taken from the grid. Because we don’t have that particular information, we’ve been using the information we do have about contracts to recover the data on how these contracts are indexed, and whether there’s any difference between these. And although so far we’ve found that there’s not much difference in the way contracts are being indexed, we’re still researching why in some places companies compete much more aggressively in the auctions than in the bilateral market. And here we found very interesting and encouraging results. Here you have the price of the contract andhere there’s supposed to be the opportunity cost for that contract. In the case of the auctions, remember, you sign that contract or bid today, but you only expect to start delivering energy three years from now, so the expectation you need to have for that, meaning what your opportunity cost will be, is described here. Here we have the interaction term between that opportunity cost and whether you’re a large buyer or a retailer in this bilateral negotiation. And here’s the length of the contract, which could be for five or ten years or even just for two (note that some contracts are very high because they were for two years, and we’ve controlled for that here); and finally you also have the size of the contract. These, as I’ve said, are very preliminary results. But what we’re finding is that at the auctions firms are bidding close to what we think is their opportunity cost (here we’re talking about a figure that’s just 3% above that). Of course we may be wrong and we still have work to do, but that’s what seems to be emerging: that in auctions, firms compete reasonably close to their marginal costs while in the bilateral trading institution, they don’t compete as close as that (we’re finding a 45% margin so when you add these two together, that gives you 1.48 or a margin of 48% above your opportunity cost). To return to the question of whether generators behave differently depending on which market they’re in, the answer is clearly that they do. As our research continues, I’d argue, we’re going to go on seeing these differences in numbers; the question is, how notable will it be? In the end, however, this disproves the premise put forwards by Chilean regulators back in the early 1980s when it was thought that simply allowing big buyers to procure the energy with the generators would produce very competitive results and that those numbers could then be used to set prices for the residential sector. It suggests that auctions should have been used from the very beginning. That, in the end, would be the lesson here: that auctions are good because they bring all the players together in one place and the players compete much more aggressively than when you have bilateral negotiations. That, essentially, would be the message these figures are sending us. But as I said, they are preliminary data and we’ll see more about where they stand as our research continues. Thank you.
11
CÀTEDRA DE SOSTENIBILITAT ENERGÈTICA DE LA UB CÁTEDRA DE SOSTENIBILIDAD ENERGÉTICA DE LA UB CHAIR OF ENERGY SUSTAINABILITY AT THE UB
