TO CONTACT: +44 (0) 20 7376 4418

Casenote: December 2018

Econometrics rejected in BritNed cartel case

BritNed v. ABB [2018] EWHC 2616 (Ch) is the first English cartel damage judgment and the first to consider an econometric approach to overcharges. The court rejected the claimant’s econometric analysis as ‘too complex’ and “unspecific”. Here I look at why it was rejected and the impact the judgment is likely to have on the future use of econometrics in competition cases.

The power cables cartel
BritNed is a follow-on damages action based on the European Commission’s cartel decision Case AT.39610 – Power Cables. This found that ABB was a member of a global cartel tendering for the supply of extra high voltage power cable projects during the period 1999 to 2009. ABB successfully bid to supply a submarine cable to BritNed’s electricity interconnection project between the UK and the Netherlands. BritNed sued ABB for alleged overcharges, lost profits and compound interest (the last two failed). The claimant used econometrics to estimate an overcharge of around 22% suggesting damages of €61.3m. The court rejected this evidence and awarded €13m (later reduced to €11m) using a cost-based method (see November Casenote).

The claimant’s econometric model.
The claimant’s econometric evidence consisted of a single during-and-after price regression. This was estimated using the ordinary least squares (OLS) technique. The data consisted of 92 ABB submarine and underground cable projects for the period 2001- 2016. It therefore did not cover the first two years of the cartel. Several ‘control variables’ were used to account for costs, the difference between underground and submarine cable projects, demand and a time trend. The cartel effect on prices (contract value) was captured by a dummy variable for the cartel period/projects. The comparator to estimate the overcharge was the competitive period after the infringement.

The Court’s approach to statistical evidence
As is usual the experts exchanged several reports prior to the trial dealing with matters raised by the claimant’s evidence and responses to points raised by the defendant’s expert. In addition, they produced a joint statement which listed the areas of agreement, disagreement and reasons for disagreement, and a further joint statement after trial addressing questions put them by the judge. Unusually the judge requested a half-day “teach-in” before the trial so that ‘each expert could provide a neutral explanation, under oath, of their working methodology.’ This went through the basics of multiple regression analysis and the Stata statistical programme used by the claimant. It raised procedural concerns that evidence was being given before trial to the judge without the usual safeguards. However, these initial fears were allayed as the issues raised in the teach-in were tested at trial. 

What the court said
The court looked closely at the econometric evidence. The judgment shows a judge increasingly critical of the econometric evidence, which he finally rejects in uncompromising language.
          Much space was devoted to two preliminary issues – data and costs:

Data. Two related issues were raised about the data – the inclusion of ABB’s underground cable projects and the small sample size. The court concluded that ABB’s underground and submarine projects differed significantly so that the former should have been excluded from the econometric analysis. This halved the sample size and exacerbated the next problem. The sample was too small and meant that the cartel dummy had a large standard error and lacked precision. As the judge observed ‘the confidence interval of the model is scarcely impressive’. The point estimate was around 22% but there was a 95% chance that the true value lay between 0.32% and 39% overcharge implying losses of €885,000 to €108.7m. This ‘shocked’ the judge who concluded that it was ‘an indicator that the model is not producing useful outcomes such that I can rely upon.’ The court’s treatment of statistical significance requires more consideration but is beyond the scope of this short note.

Proxy costs. The claimant’s expert declined to use ABB’s costs as an explanatory variable in the regression because in her view they ‘lacked transparency and consistency’, there ‘might be biased reporting during the cartel period’ and costs may have been inflated by the cartel. Instead she used ABB’s copper and aluminium input prices as a proxy for its project costs as these would not have been affected by the cartel. The court disagreed – ABB’s cost should have been relied on and the claimant’s costs proxy was ‘insufficiently aligned with the actual – highly individual – costs of submarine projects’ [417]. This was a strange position to take since the court subsequently accepted that ABB’s costs were affected by the cartel and excessive and which it used to quantify the damages (see November Casenote).

Sensitivity tests raised further doubts
Econometric evidence should be subject to sensitivity tests to ensure ‘robustness’. These look at whether the estimated overcharge alters markedly with the exclusion of key control variables, different time periods and/or the specifications. If it does then this indicates problems with the regression estimate. Both experts carried out sensitivity tests although the judgment focuses on those of the defendant’s economist. These involved excluding in turn and separately cartel projects other than the BritNed project, underground cable projects, the time trend and ‘order backlog’ variable. With one exception these reduced the overcharge estimate and rendered it statistically insignificant. This by itself was not a matter for concern. As the judge commented [379]: ‘If the parameters are material … their removal from the model will make a difference’ (emphasis in original).
          Nonetheless the sensitivity tests set in train questions which undermined the probative value of the econometrics. For example, excluding underground cable projects from the data halved the sample size, increased the overcharge to 27.7% but this was ‘statistically insignificant’, and altered the coefficients of several control variables rendering the time trend insignificant. The ‘overcharge’ coefficient should not have altered much if underground and submarine cable projects were sufficiently similar.

The problem of averages
Then came the killer blow. The judge [418] said ‘the fragility of the model is in large measure hidden by … [the] use of averages.’ A regression using a cartel dummy variable estimates an average overcharge which the claimant’s expert used ‘to compute the overcharge for the specific case, the BritNed project’. When the model’s parameters were applied to individual submarine projects to generate predicted overcharges there were large differences – some small, some negative and others massive. Further, the econometric evidence found that ABB had on average overcharged by 22% while the judge concluded that the documentary and witness evidence suggested that ABB had competitively priced the BritNed project.
          As the judge [421] commented ‘given the bespoke and unique nature of these projects, I find that an overcharge calculated by a model that is explicitly averaging across multiple projects to be an inappropriate one’. This was a valid criticism given the highly differentiated nature of ABB’s projects. Lumping them together and suggesting that the average overcharge was applicable to any one project is hard to defend.
          The court’s rejection of averages posed an unappreciated problem for the way it calculated damages. Simply, if ABB’s projects were so ‘bespoke’, then one could not compare the BritNed project’s margin and costs with the averages of these for ABB’s post-cartel projects as was done by the court, and indeed the defendant (see November Casenote). At least the claimant’s econometrics sought to adjust for the differences which the court failed to do.

While one can criticise aspects of the judgment, and whether the court’s preferred approach satisfied similarly exacting standards it applied to the econometrics, it could not be said that the econometric evidence was unreasonably dismissed. Given the nature of ABB’s projects it was questionable whether an econometric approach was best suited and should have formed the centre piece of the claimant’s approach to quantification. Apart from the statistical issues raised in the judgment, which will no doubt be rehearsed before the Court of Appeal, the approach floundered on the use of a statistically estimated average overcharge which was then applied to a specific project. An average estimate will invariably be ‘wrong’ for any one ‘bespoke’ project.
          BritNed is not a setback for the econometric approach, but simply the rejection of the claimant’s econometric evidence. But it does point to the need to treat statistical evidence as complementary to and ensure that it is consistent with the documentary and other evidence. The problem in BritNed was that the other evidence did not enable the court to quantify damages by any approach which was credible and not itself flawed (as argue in my earlier Casenote). Econometrics was perhaps the best of the worst approaches.

© Copyright Cento Veljanovski, December 2018. (The author was not involved in this case). An updated version of this Casenote was subsequently published in eCompetition news 2019

Search this website
Close Menu