Normally, merger simulation - formally modelling what might happen (particularly to prices) if competitors merge - is done before the event by competition authorities (and also, sometimes, by economics consultancies, generally in support of the merging parties). What the two authors of this article have done, however, is turn the process on its head: instead of asking, before the merger, what post-merger effects does the model predict, they ask, after the merger, what model would have best predicted the effects that actually happened.
Usually, this isn't doable. If a merger is cleared, and it doesn't substantially lessen competition, generally there are no obvious price effects to see. Fortunately for the authors - but rather unfortunately for the Swedish competition authority - they've got data for a clearance that went badly wrong. Prices were jacked up substantially and immediately post-merger, and to some degree by third party producers.
Here are the facts in approximate brief. Analgesics are painkillers: there are three main ones, paracetamol, ibuprofen and aspirin. The only two makers of paracetamol in Sweden applied in late 2008 to merge, and the competition authority allowed it in April 2009. The regulators relied on a broad market definition (the other two painkillers would be good substitutes, constraining any rip-off on paracetamol) and on the prospect of greater competitive constraints following an adventitious deregulation of the pharmacy sector (up to late 2009 there had been a state-owned pharmacy monopoly).
In fact, the price of paracetamol went up by some 40% immediately post-merger, and the aspirin makers cashed in too, with price rises from 7% to 18%. Ibuprofen prices didn't change much (no, I don't know why, either). Here's the graph ('ASA' is aspirin).
Clearly, the painkillers weren't in fact good substitutes for each other (the facts speak for themselves, and when they ran their models the authors also found low cross-price elasticities) and the paracetamol makers were able to coin it. There were lucky, in that an anti-drug-overdose measure was implemented around the same time, and they were required to sell 20-tablet packs instead of 30-tablet ones: they were able to smuggle in higher per-tablet prices by not reducing packet prices proportionately. The authors allow for the fact that it's more expensive to make smaller packs, but even so that would have accounted for only around 15 percentage points of the 40 percentage points price increase, leaving 25% as pure lower-competition gravy.
The authors were also able to go a long way with their main research interest - which models of producer behaviour and consumer demand best fitted the facts (answer: none perfectly, but some pretty impressively).
Okay, that's the specifics of their work, but I think their research also makes some more general points.
First, it suggests that (at least some) merger simulation models are well worth running as part of the clearance process. Our own Commerce Commission used to have a formal model, but it may have been put out to graze: a search today of the website for "merger simulation" or "Bertrand" came up empty, and the five "econometrics" results weren't relevant. This Swedish study, however, suggests there's a dance in the old dame yet. I wouldn't push the argument too far: as Stephen King, at the time one of the ACCC Commissioners, said in 2005 in 'The use of empirical methods in merger investigations'
That may be true, but it's beginning to look to me like a low-ball estimate of the potential value of merger simulation modelling.Because of its complexity and sensitivity to particular assumptions, merger simulation is generally contentious and, at best, provides ‘back up’ input for a more complete merger analysis
Secondly, I think it says something about the potential use of more econometrics in competition and regulation analysis more generally. So far, it's been a bit of an uphill struggle:
Lawyer: So, your model rests on quite a specific set of assumptions?But we're in a new world of big data where crunching the numbers with better tools is getting easier and more reliable. It's time, I reckon, to fire up more models, more often. As I've said before
Economist: Yes, but...
Lawyer: And if those assumptions do not hold exactly, the results may not be reliable?
Economist: No, but...
Lawyer: And they haven't held exactly, have they?
Economist: No, but...
Lawyer: No further questions of this witness, m'lud.
we may be getting closer...to being able to do a better job of taking a more robust empirical approach to measuring things like demand curves, and own- and cross-elasticities of demand. If, using things like scanner data, improved econometric methods, sophisticated consumer choice testing, and clever analysis of 'natural experiments' - what happened, say, after a fortuitous interruption to one source of supply - we can get a more scientific handle on the extent to which products are or are not substitutes for each other (and so are or are not likely to be in the same market), why wouldn't we use that information to derive empirically grounded market definition?