1. Introduction
So far, Mexico organized three different longterm electricity auctions. Those auctions were characterized by extremely low bids which caught many outside observers by surprise. The Italian firm ENEL, for example, offered electricity in the third auction at a worldrecord low price of 17.79 USD per MWh.
In this note, we argue that computing the average winning prices is actually far from simple. In particular, we stress the following:
i. Careful attention should be paid to the treatment of inclusive bids, and
ii. average winning prices should be computed using hedonic regression techniques.
We apply our methodology to the second and third longterm electricity auctions and we report the average winning prices for 1 MW of power and for 1 MWh of cumulative energy (CE) + 1 clean energy certificate (CEC).
2. Why is it so hard to compute unit prices in the Mexican electricity auctions?
Computing average winning prices in those auctions is complicated for the following three reasons.
1. Inclusive bids. Imagine you want to buy a pair of shoes and consult various shoe producers. To each shoe producer you ask: “How much do you charge for the left shoe and how much for the right shoe?” (That question obviously makes no sense but just imagine that all shoe producers take your question seriously.) Suppose also that all shoe producers want to earn 200 pesos for each pair of shoes sold. The first shoe producer might then answer that she wants 100 pesos for the right shoe and 100 for the left shoe. A second producer might come up with a completely different answer. He could ask, for example, 199 pesos for the left shoe and 1 peso for the right shoe. A third producer might ask 1 peso for the left shoe and 199 for the right shoe. This example illustrates that individual prices make no sense in the context of complementary goods.
A similar thing happens in the electricity auctions. To illustrate, consider the following example:


























In the example, two bidders (Bidders 206 and 217) submit two different bids. Bidder 206 indicated that her two bids are mutually inclusive, i.e. CENACE cannot declare Bid 1 as a winning bid without doing the same for Bid 2 (and vice versa). Bidder 217 also indicated that his two bids are mutually inclusive. Observe that Bidder 206 offers CECs at a cheaper price than CE, while the contrary occurs for Bidder 217. Bidder 206 actually sells one CEC at a price of 210 pesos, while Bidder 217 offers CECs at a unit price of 326 pesos, a 55% increase!
The following table, however, shows that the difference between bidders 206 and 217 is altogether not that large.



















The previous table shows that Bidder 206 computed the total price of her combined bids using a tariff of 517 pesos per (MWh + CEC) while Bidder 217 used a tariff of 562 pesos per (MWh + CEC). Bidder 217 is thus “only” 8.6% more expensive than Bidder 206.
Example I shares thus the same implication as our previous example with the shoe producers: In the presence of inclusive bids (or in the presence of complementary goods), one should not pay attention to individual prices. In any statistical analysis, one should thus add up all inclusive bids —as we did in our previous table— to avoid reaching erroneous conclusions.
2. Some bids include power, cumulative energy and clean energy certificates. Consider now the following example:





















The first two rows in the above table represent two bids submitted by a bidder in the second longterm auction. (The bidder did not win any of them.) In the first bid, the bidder offers slightly more than 90,000 MWh of CE at a price close to 45 million pesos. In the second bid, she offers power and CECs at a price close to 32 million pesos. The bidder indicates that both bids are mutually inclusive. As argued previously, both bids should thus be added up. (This is represented in the last row of our previous table.)
Our bidder thus asks a price close to 77 million pesos for three different quantities of three different products. Ideally, we would like to decompose those 77 million pesos in three different components:
3. The quantity of CE can be different from the quantity of CECs. This point is actually related to our second one.
Consider the first longterm electricity auction. After adding up all inclusive bids, one realizes that that auction counted “only” 16 winning bids. None of the winning bids included power. Twelve of the winning bids offered exactly the same amount of CE and of CECs. The remaining four bids featured a very small difference between both quantities. As a matter of fact, the difference between the amount of CE and the number of CECs bought is less than half a percent. (The number of CE bought is actually slightly higher than the number of CECs.) In other words, in that auction it is very reasonable to assume that the supply of CE equals the supply of CECs. Computing the average combined unit price of one MWh of CE and one CEC is then very simple: just add up the prices of all winning packages and divide that number by the amount of CE bought by CFE. This procedure actually yields an average price of 47.11 USD per MWh and CEC.
In the subsequent auctions, however, small differences started to appear between the supply of CE and of CECs. To be more precise, in the second auction the number of CECs sold is 4% higher that the number of CE. In the third auction this percentage actually goes up to 8.4%. Stated differently, there is a tendency in those auctions to gradually increase the relative supply of CECs. We suspect this tendency to continue in the fourth longterm auction that will be held later in the year.
Consider now a typical winning bid in the second or the third longterm electricity auction that does not contain any power. To compute the combined price for one MWh of CE and one CEC, one cannot simply divide the total price of the package by its supply of CE; one should instead subtract some amount of the total price to reflect the fact that the package contains a larger supply of CECs. But how much should be subtracted from the total price?
3. A hedonic regression approach
To compute the average winning prices, we first add up all mutually inclusive bids. Next, we perform the following regression:
Price Package = Β_{power} × Power + Β_{CE} × CE + Β_{CEC} × CEC + ε,
where:
We applied our methodology to the second and the third longterm electricity auctions. (As mentioned previously, the average winning price in the first electricity auction can be easily computed.) The table below summarizes our main findings:





(67,868) 
(27,457) 

(12.27) 
(10.07) 
The following table computes the average prices in dollars. For the second auction, we used an exchange rate of 19.152 pesos per dollar. For the third auction, an exchange rate of 19.1474 was used. Average unit prices in the Mexican electricity auctions (in USD).










4. Final observations
SENER, the energy ministry, reported that the average winning bid for one MWh of CE and one CEC in the first auction was equal to 47.78 USD. In the second one, this combined price decreased to 33.4 USD. In the third one, the combined price decreased even further to reach 20.57 USD. Our estimates are very close to the ones reported by SENER, except for the prices obtained in the third auction. The difference between our estimates, 1.51 USD (= 20.57 – 19.06), is statistically significant.
Suppose a bidder considers including 10 MW of power in her bid in the next auction. Our estimates suggest that she could then raise the price of her package by 7 million pesos without affecting the competitiveness of her bid. (This, of course, assumes that the price of power will not decrease further in the fourth auction.)
5. Bibliography
Ernst Berndt, The Practice of Econometrics, Classic and Contemporary, AddisonWesley 1991, 702 pages.
6. Acknowledgments
This note was written by Dr. Nicolas Melissas. Nicolas gratefully acknowledges Veronica Irastorza and Laura Juarez for helpful comments on an earlier draft.