Listen to the podcast:
Collusion among companies in the form of cartels is generally viewed as detrimental to the marketplace. The Antitrust Division of the U.S. Justice Department has uncovered some cartels, but it is believed that many more go undiscovered. Authorities typically examine the cartels they’ve uncovered to learn about cartels yet to be discovered. But is their approach flawed?
A research paper by Wharton business economics and public policy professor Joseph Harrington titled, “What Can the Duration of Discovered Cartels Tell Us About the Duration of All Cartels?” examines discovered cartels to see whether they accurately describe the actual latent population and duration of all cartels. Knowledge@Wharton recently spoke with him to talk about his findings.
An edited transcript of the conversation follows.
Knowledge@Wharton: Tell us about your paper.
Harrington: Well, it starts with the issue of collusion. And probably [a description of its harmful effects] was best stated by the recently-deceased [Supreme Court] Justice Antonin Scalia, who referred to collusion as “the supreme evil of antitrust.” Collusion is all about the fact that firms — which we count upon to compete for the business of customers, and doing so results in lower prices, better products and the like — in some industries, decide that [competition] leads to profits that are too low. So they engage in unlawful coordination of their behavior. Instead of driving prices down in order to get customers’ business, they decide, “Well, how about we all set artificially high prices, share the market, and we’ll all earn high profits.”
And so collusion is a real challenge [to a properly functioning marketplace]. And it’s become particularly so in the last several decades. I would say 30 years ago, if you asked an official of the Antitrust Division of the Department of Justice whether there are any global cartels, other than types like OPEC, they would have said no. But the fact is, we have observed many global cartels, as well as many domestic cartels, in the last couple of decades.
“If cartels tend to largely vary in terms of the likelihood of collapse, what we find is that the measured duration of discovered cartels is an overestimate of the true duration.”
But a challenge in terms of understanding cartels [is this]: How many unlawful cartels are there? How bad are they? How much do they raise prices? How long do they last? We face a challenge common to criminal behavior, which is, these criminals want to hide themselves. So what we observe are just those cartels that are unfortunate enough to be discovered and convicted.
So the challenge [becomes]: “Well, we know how many discovered cartels are out there. We know how long they last. We know how high a price they set. But is that representative of the actual latent population of all cartels?”
Where this specifically manifests itself in terms of a policy challenge is, suppose the Antitrust Division puts in place a new program and they want to know, “Is it proving to be beneficial? Is it actually helping to reduce the presence of collusion?” Well, all we can observe is what’s happening to the population of discovered cartels. So if, for example, we observe more discovered cartels, is that because the program is working to make discovery more likely, and that’s good? Or maybe it’s counterproductive. Maybe what it’s doing is actually creating more cartels, and that’s why there’s more discovered cartels.
That’s the starting point to this research, which is trying to determine what it is we can learn about the underlying latent universe of cartels from those that we actually discover. More specifically, what the research is concerned with is the issue of the duration of cartels. There have been many studies that have measured how long discovered cartels last. In these studies, the average duration ranges from five to eight years, depending on the study. And the issue is, “Well, is that a good proxy for the duration of all cartels?” What this research is trying to provide is a method for measuring how much bias there might be from looking at discovered cartels, with regard to the universe of cartels.
Knowledge@Wharton: What are some key takeaways of the paper?
Harrington: To answer that question, let me say that the paper makes two contributions. First, it puts forth a theoretical framework to think through these issues and understand, “Well, when will bias occur? In what direction?” And the way it does this is to construct a model of the birth, death, and discovery of cartels.
If you read many papers in the literature, some economists will say that, “Well, the measured duration of discovered cartels is an overestimate of the actual true duration.” And some will say it’s an underestimate. And so what we want to do is put forth a theoretical framework where we can say precisely, when is it an over- or underestimate? What drives that?
… Let us think about cartels as differing in two key characteristics: what’s the likelihood of collapsing, let’s say, within a year; and what’s the likelihood of being discovered in a year? If cartels tend to largely vary in terms of the likelihood of collapse, what we find is that the measured duration of discovered cartels is an overestimate of the true duration. And the reason is that, if you have a cartel that’s really very stable, with a low chance of collapse, it just hangs around for a long time and has plenty of opportunities to be discovered. Cartels that are short-lived will tend to die and avoid discovery.
So the bias in that case will work toward saying that, “Well, if we measure, for example, the average duration of discovered cartels to be, let’s say, six years,” that probably is an overestimate. Now, there are other assumptions you can make, whereby you’d get bias going the other direction.
So one contribution and takeaway is to be able to frame these issues, so we understand, “OK, what drives bias?” The other contribution, and more where I think the key takeaway is, is to then use this framework to get a measure of the extent of this bias. How big is it?
“The other contribution … is to then use this framework to get a measure of the extent of this bias. How big is it?”
We take this framework, and we use data on convicted cartels by the Antitrust Division, from cartels that were born from 1961 to 1984. First of all, you could just measure the average duration of these convicted cartels, and it’s 5.8 years. But we could use this framework to then get a sense of how big is the bias with regard to the duration of all the cartels. Also those that are undiscovered.
And what we find — and I think one of the surprising results – is that the bias, actually, is not that large. We find a high likelihood [that] the average duration of all cartels probably ranges from about 5.2 years to 6.8 years.
Knowledge@Wharton: What are some practical implications of your research?
Harrington: Probably the most substantive one is with regard to evaluating policy. Now, I think there’s been a lot of policy innovations by various competition authorities. One of the big challenges is determining, “Are they helping in the fight against cartels? Are they resulting in fewer cartels? Reducing the overcharge? Resulting in lower duration?”
As I said, the challenge here is that we don’t directly measure those things. So it’s not like with some types of crime, where we can measure the crime rate, because all the crimes are reported. With the case of cartels, those who are harmed don’t necessarily know that they’ve been harmed. They don’t know they’ve paid excessively for these goods.
So that’s a real policy challenge. And to be more concrete about this, probably the biggest innovation in policy in fighting cartels in the last 30 years has been the revision of the Corporate Leniency Program by the Antitrust Division. This is an age-old idea [to deal] with conspiracies, which is if someone from the conspiracy comes forward, cooperates with the authorities, they’ll be absolved of government penalties. This was a program that was in place starting in 1978, but it had certain design flaws to it, and was rarely used. In 1993, it was revised and made structurally sound. And immediately, the leniency applications started coming in.
So it’s certainly been successful as measured by the number of leniency applications. It’s been adopted by many other jurisdictions. Probably more than 50 countries and unions have a leniency program. But still, the question is, “Is it affecting the actual population of cartels?” And that’s a difficult question.
One of the things we want to do here with this kind of approach is to provide a method to measure the impact of a program such as the leniency program, to specifically address the question of, has it, for example, resulted in cartels being of shorter duration? That would be an immediate benefit from it. And so that’s something we plan to do in the next step of this research project. We’ve used this model to measure the underlying duration of cartels with data prior to the revisions in the leniency program. A new data set has recently become available, and that will allow us to [track] cartels that were born and [those that] died after the advent of the leniency program.
We can redo the analysis and see, “What has happened to the duration of cartels before and after the leniency program?” We won’t be able to tell any kind of causality story, but I think it will provide some needed data to speak to the issue of, “What impact are these programs having?”
Knowledge@Wharton: What conclusions, if any, surprised you?
Harrington: The biggest one was that there wasn’t as much bias as I would have expected. That’s been a running concern of economists and policymakers in this context. The bias is definitely there. We definitely find evidence of it. But it’s not as large as one would have thought. So let’s put it this way. I think some of the conclusions drawn based upon what were known to be biased estimates are probably robust, because of the fact that we found that the bias isn’t actually that large.
Knowledge@Wharton: What sets your research apart from prior work in this area?
“One of the surprising results is that the bias, actually, is not that large.”
Harrington: This is a topic people talked about and mentioned saying, “Our data’s probably biased, because it’s just discovered cartels. And those cartels could be very different from the universe of cartels. You could easily imagine that, well, we’re just discovering the ones that are very ineffective, and that’s why they’re being discovered.” Or as I said before, we could be discovering the ones that are really stable and effective, and so they’re around for a long time, with lots of opportunities to be discovered.
That was kind of recognized and mentioned. And then people just went forward and did their work as if the data was not biased. This was really the first study to say, “Let’s see what we can actually say about the extent of the bias? And is it really that large? And to what extent can we compensate for it in our conclusions?”
Knowledge@Wharton: How will you follow up this research?
Harrington: There’s going to be a follow-up empirical analysis, using data from the post-leniency program era, to get some sort of assessment about whether duration has been impacted by the leniency program.
I think where one could also make advances is in terms of the theoretical framework. Right now, the framework itself is kind of sparse in terms of how it models the birth, death, and discovery of cartels. But there’s a developing literature in economics on how to model birth — what determines whether or not a cartel is created? What inspires managers to say, “Well, let’s not compete anymore. Let’s collude.”
[Also, we want to look at] what causes cartels to collapse. There’s a rather large body of work on that. And less so in terms of what leads to discovery. But I think trying to take into account a richer structure and to embed them within our framework, so that we can tell a richer story about the life cycle cartels, from when they’re born to when they die.