Division Rationale and description of the econometric analysis

The econometric analysis undertaken by the Commission is designed to create benchmark prices for contracts. These prices result from a regression analysis carried out on the features of the contracts in the dataset. As a first step, the regression analysis allows a benchmark price to be constructed on the basis of the characteristics of the contracts in the dataset (‘in-sample predictions’). As a second step, the results of the regression analysis are used to predict a benchmark price for the contracts investigated, taking into account the characteristics of the contracts (‘out-of-sample predictions’). In the regression analysis, the variation in prices among the contracts in the dataset is explained with reference to the following characteristics: quantity purchased, supplier identity and year dummies. The ‘contract duration’ and ‘buyer off-take profile’ variables are not included since they are not statistically significant.

The econometric analysis rests on the principle that there are a number of drivers of prices, such as quantities. It would be misleading to compare the prices across different contracts without considering these drivers. In particular, whereas the contracts in the dataset have certain characteristics in common with the contracts under investigation (bilateral negotiation, bespoke conditions, quantities above 150 GWh per year, etc.), no contract in the dataset has exactly the same quantities, or was concluded for the same duration and with the same entry into force, as the contracts under investigation — a factor that may justify price differences. For example, two contracts under investigation concern supplies of […] GWh/year (Alro 2007-2010) and […] GWh/year (Energy Holding 2007-2008). In the meantime, the contract with the highest quantity in the dataset out of the 114 price/contract data observed between 2007 and 2010 did not exceed 1 400 GWh/year. It is doubtful whether a mere comparison of prices for these contracts could allow any valid conclusion to be drawn as to whether the contracts under investigation were in line with market prices. Alternatively, it could be concluded that these two contracts could not simply be compared with other market contracts. The rationale of this quantitative exercise is therefore that, once a number of external factors are taken into account, the prices across different contracts can be better compared. In the absence of normalisation, only perfectly identical contracts could be compared meaningfully.

It should be noted that this empirical exercise is not aimed at estimating a causal relationship between prices and certain external factors. For example, estimating a causal relationship between certain factors and prices would require the risk of endogeneity to be taken into consideration: that is, the risk that a causal variable (e.g. the quantities) might itself be influenced by the explained variable (e.g. the price), due to omitted variables or simultaneity bias. The purpose of the quantitative exercise is to ‘normalise’ the prices across different contracts to enable a better comparison to be made between them. This normalisation is necessary in the absence of perfectly identical contracts and supply characteristics.

The regression analysis reflects the main features of the bilateral contracts at issue:

• including the quantity variable in the regression reflects the fact that prices are in general lower when quantities purchased are higher^f1,

• including the supplier identity reflects the fact that some suppliers have characteristics which would allow them to establish prices differently from others, taking into account that contract supplies from virtually all electricity suppliers active in Romania are included in the dataset,

• including year dummies reflects the time dimension and the possible change in market conditions from one year to another.

As regards the first step of the empirical analysis, the results of the regression analysis of the dataset contracts are reported in Table 1 below.

As regards the second step of the empirical analysis, the Commission has identified a benchmark for every year and has then tested the position of the contracts relative to this benchmark in order to conclude whether the prices charged by Hidroelectrica were lower or higher than the modelled benchmark price. The following steps describe in detail the methodology used to determine the benchmark:

Firstly, for each contract of the dataset, it was calculated how far the actual price departed from its corresponding benchmark price, calculated on the basis of the regression and of the characteristics of the contract;

Secondly, the most-downward-diverging contract (‘MDD’) was identified; this was the dataset contract whose observed price departed most from its own corresponding benchmark price (in absolute terms). There were reasons, very conservative as they were, for choosing the MDD contract, involving a range of variation below the central estimated value of the benchmark price. Firstly, the econometric model does not fully explain the observed price in the dataset, and the single estimate of the benchmark price was provided with a confidence interval and a margin of error above or below the estimate. Secondly, deviations from a single possible price exist on the real market; the MDD contract, which stems from a contract based on market conditions (see recitals 36 to 38) supplies quantified information about the possible extent of such deviations and provides a market-based range around the calculated benchmark price.

Thirdly, the price difference in relation to the MDD contract is used to separate contracts below the benchmark price from contracts above this:

• if a contract has an observed price below its corresponding benchmark price and if the price difference for this contract is larger than the MDD price difference, then this contract is considered prima facie not to be market-compliant,

• otherwise the contract should be deemed to be market-compliant.

The table below presents detailed results from the regression analysis of the sample of contracts from the dataset. The regression explains 74 % of the variations in the data, which is a relatively good fit. The coefficient estimates presented in the table below are used at a second stage to predict the ‘benchmark’ price for the sample of contracts under investigation (‘out-of-sample predictions’).