The ACCC will not be appealing the Full Federal Court decision in the Metcash-Franklins merger. I think that is the correct call. A further appeal would be unlikely to reverse the decision and, if the High Court took the same approach as the Full Federal Court, would offer little guidance on underlying principles of comeptition analysis for mergers.
The decision, however, leaves two key issues unresolved.
The first is clearly concerning the ACCC and is discussed in their media release. Analysing a merger means economic prediction – how will the market behave with or without the merger? The ‘without’ or ‘counterfactual’ analysis is often broken into two steps:
- What are the likely market outcomes if the merger doesn’t go ahead: and
- What is the state of competition with these market outcomes.
In simple merger cases, the likely market outcome in the absence of the merger is the status quo. This is easy to analyse as we have historic evidence on the nature of competition. But what if the status quo is not relevant, as in the case of Franklins, where a sale – either as a whole or piece-by-piece – was the counterfactual? In this case, the counterfactual analysis can be difficult and controversial.
What is the ‘probabilistic threshold’ for counterfactual analysis? A merger is illegal if it is ‘likely’ to substantially lessen competition (SLC). The courts have (at least) two alternative views of the word ‘likely’. The first is that it means ‘more likely than not’, so the hurdle is that a merger must have a greater than 50% chance of substantially lessening competition relative to the counterfactual(s). The alternative is that the SLC is a ‘real chance’ if the merger occurs. The High Court in New Zealand said that a 30% probability represented a ‘real chance’ in the Warehouse matter.
The Full Federal Court has left it completely unclear what probabilistic standard should apply when considering if a merger is illegal. One judge believes that it is clearly a ‘real chance’ threshold. A second argues that it is ‘more likely than not’. The third is silent. Hence the ACCC’s response. It is going to continue using the ‘real chance’ approach based on significant precedent, until the Court clarifies its position.
More worryingly from the perspective of someone with some statistics, the Court wishes to apply these probability hurdles separately to each step of the counterfactual analysis. So should the ACCC have to prove a ‘specific’ counterfactual is ‘more likely than not’ or a ‘real chance’? And then given the counterfactual, should the ACCC have to show that an SLC is ‘more likely than not’ or a ‘real chance’ given that counterfactual?
This is a real problem. To see this, suppose the Court decided that the overall test was that the ACCC has to show a SLC is more likely than not. Further suppose that the Court applies this analysis separately to each leg of the test. So say that a specific counterfactual is 60% likely (so more likely than not) and, given that counterfactual, an SLC is 70% likely. Is the merger likely to substantially lessen competition with a probability greater than 50%? Not on the basis of this counterfactual alone. The probability of an SLC is 42% (i.e. 0.6 times 0.7).
Of course there could be other counterfactuals, with associated conditional probabilities of an SLC. A simple application of Bayes’ Law will get the overall probability of an SLC. But this is not what the Court is discussing. They apear to be approaching the SLC test in a way that makes little statistical sense.
One way out of this is for the Court to realise that proving ‘specific’ counterfactuals is not really relevant. Counterfactuals in merger analysis are not externally imposed ‘states of the world’. Rather, when thinking about a merger, analysis tends to group future possibilities as ‘situations where an SLC is highly likely’ and ‘others’. In the Metcash case, the relevant division was simple:
Counterfactual One: Someone (other than Metcash) buys Franklins and proceeds to operate it as a competitive grocery wholesaler to independent supermarkets; or
Counterfactual Two: Franklins is not sold to Metcash and is broken up.
Given these two alternatives, the conditional probabilities are simple. If the first counterfactual is relevant then an SLC is (close to) certain. If the second counterfactual is relevant then an SLC is not going to arise with a sale to Metcash. So the real question for the Court is: ‘Is it likely that someone will buy Franklins and use it to compete vigorously in the supply of wholesale groceries’? This can be judged against either a ‘more likely than not’ or a ‘real chance’ threshold.
In summary, the two ‘steps’ of the counterfactual analysis are intimately connected. Merger analysis needs to recognise this connection. It can do this by a full application of Bayes Law over all possible counterfactuals, or by grouping counterfactuals into those where an SLC is highly likely and those where it is unlikely. Given that the evidence is rarely available to carry out a full analysis, I prefer the second more pragmatic approach. However, currently the Court does not seem to do either of these alternatives.
The second unresolved issue from the case is the determination of vertically separate markets (called ‘functional markets’). The Federal Court was consistent in this matter and considered that there is a market for groceries that included both retail and wholesale levels. It made this conclusion on the basis of competitive constraint. While I have some concerns about how the Court considered those constraints, the real issue is that this approach, while not unusual, seems at odds with other Court decisions, particularly those in Part IIIA access matters. So the Court failed to clear up an existing ‘tension’ in application of the Act.