> Economists typically assert that cartels dissolve naturally after members cheat or become irrational. When computers are the actors, though, detection is faster and not prone to human errors or failings, making defection less likely. Automated participants can identify price changes more quickly, allowing defectors who lower prices at the expense of the group to be sifted out earlier. Given this dynamic, participants have little incentive to either “cheat” the group or to leave it.
What is the penalty for defecting from the group to clear the market? An actor gets kicked out of a group in which he was a voluntary member?
Suppliers take a risk by price-fixing in that they may be underutilized, or be left with excess supply. Breaking even occurs when a few customers paying the collusion price net the same income as many customers paying a spot price. Clever suppliers can profit, but all actors are still independent. The only recourse for clearing the market is being left out of future collusion, which means nothing if you're clearing 100% of your supply and the adversary is clearing 0. The only choice of all actors is to compete for satisfaction.
Your analysis is correct if we optimize our algorithm to maximize the profit from its most recent trade.
But, as with the iterated prisoner's dilemma, what makes a successful strategy can change when a game is played repeatedly. Imagine a genetic algorithm that comes up with a tit-for-tat strategy when playing the iterated prisoner's dilemma. Now imagine a similar genetic algorithm applied to automated trading. Imagine a market in which a large fraction of the trading is done by such algorithms.
I think it is quite plausible that a cartel could form more or less spontaneously, without any conscious encouragement in that direction from human programmers. (But hey, maybe you have a good argument that I'm wrong.)
"Economists typically assert that cartels dissolve naturally after members cheat or become irrational."
Really? Most economists I know acknowledge OPEC still exists and is still a functioning cartel. And they acknowledge that many economists predicted otherwise when it formed and were wrong. I'm pretty sure even those who made such predictions (Milton Friedman from memory for one) have acknowledged they were completely wrong.
"As everybody knows..." Was how such falsehoods used to be prefaced in the bad old days in the Soviet Union, so I'm told.
ggchappell nailed it with an iterations of the prisoner's dilemma situation combined 'price signaling'. Take a single, 1 round game of pricing iPods at Wal-Mart and Target. iPods have a cost of $100 and each competitor already has bought the item. What would both competitors price at? (something a above 0 and probably logically around the cost of the product)
Now, let's take a game of 1000 rounds. Same situation, but Wal-Mart has stuck to a strategy of price-matching in each successive round. What would the price be that the competitors arrive at over time? (higher than the cost to maximize group surplus) In this situation it is no longer optimal to break rank with lower prices because the gains are short-lived and cannot make up for the effects of long-term margin compression.
I actually commented on this exact possibility on the original post about his prosecution [1]. The answer to the headline is yes, they can form cartels, and if they were smart enough, the algorithms would figure it out on their own without any human intervention that this is the most profitable way to do business. There is currently no API for bots to check with to see if they are violating the law, and it's unlikely that the author of an algorithm that ultimately decides on its own to take actions that would probably violate the law if performed by a human would be held accountable for its actions.
As you say, the problem is that, in an oligopoly, raising prices is often game theoretically optimal. And while explicit collusion may be illegal, implicit collusion (all parties noticing the best strategy independently) currently isn't. I don't see that computers add much of an interesting angle, except, as the article states, to make irrational or accidental defection less likely.
So really the law on this has always been pretty broken in theory, but maybe it worked well enough in practice. With computers that may change and the law will adapt to some other broken-in-theory system that works just well enough in practice.
>to make irrational or accidental defection less likely.
Why do you believe these behaviors are less likely? A single bug could wipe out all previous gains and incorrect assumptions could cause behavior that is later identified as irrational.
What happens is you end up with a wasteful meta-bureaucracy where competing companies and governments form committees to set prices.
I used to work for an ocean container shipping company, and due to idiosyncrasies of maritime law and the fact that these companies are considered "too big to fail" by governments, they are exempt from certain price-fixing laws in many countries including the US and Canada (but they are no longer exempt in the EU since 2008). Carriers are required to publish their rate schedules and surcharges and jointly announce rate increases 30 days in advance of the effective date.
They also share service strings, so a weekly Transpacific service may have a rotation of five different vessels from different carriers, who lease each other container space. They are officially not allowed to restrict supply in a coordinated fashion to prop up prices anymore, but they have to jointly make operational decisions to downsize or cancel service strings, which is effectively the same thing.
The result of this, along with numerous government and private bailouts, is that many of these perennially loss-making companies with inefficient operations and terrible service continue to limp along and slowly bleed shareholder value, whereas in a free market they would have been bankrupted and their assets liquidated years ago.
Though to be honest, I think anti-discrimination laws are even more broken. One reason being that price fixing has to involve multiple companies, where anti-discrimination laws can involve just one company even if there are plenty of other jobs that the employee could choose from.
> it's unlikely that the author of an algorithm that ultimately decides on its own to take actions that would probably violate the law if performed by a human would be held accountable for its actions.
You maybe right about how the courts and regulators will find in this matter.
But if you build a machine that breaks the law on your behalf and you knew this was a possible outcome then you are obviously culpable. To come to another conclusion requires an inordinate amount of mental gymnastics. And in this decade's climate, it's quite possible those gymnastics win the day.
Consider more hypothetically: someone whips up a truly magnificent algorithm to make cash for him anyway what-so-ever, not just on the exchange. The programer made no effort to preclude illegal actions because that is essentially impossible. This hypothetical super program occasionally finds the best way to make money is reselling pirated movies and web based child porn and deposing the revenue straight into the programmers account.
You watch how fast the courts ban the practice and finds him culpable without a shadow of a doubt.
No. There is an excellent work by Steve Keen that shows how simple it is for non-communicating bots under perfect competition to collude: http://arxiv.org/pdf/nlin/0411006.pdf
I wish that mainstream economists would accept that the collusion and not competition (aka price war) is the normal state of affairs and updated their school models accordingly.
"In this paper, we assume the only information a firm has is its own cost function, the decision it made
on the previous time step (increase or decrease production), and what impact this decision had on its
profit levels."
I wish that people would stop citing papers based on the abstract and complicated looking math, when the entire paper is based on a completely ridiculous assumption. That companies can collude without direct communication is not in doubt - they can just do a tit-for-tat strategy. What is in doubt is whether this is stable - new entrants can come in and use a different algorithm, play a different game. This paper does not address that issue.
It's actually not primarily based on complicated mathematical assumptions, but on the simulation and also empirical data about MR!=MC in the real world. That was his primary motivation to study this problem. See Steve Keen's book Debunking Economics if you want a bit more details.
Even if you would have an endless stream of market entrants, you would also need some companies to exit, otherwise everybody would get smaller and smaller piece of the pie. In the real world, those who come will usually set the price just below the current market value. If you think about it, it's nothing else than the Keen's model with just companies being renamed endlessly (after they make one step).
But go ahead and try to simulate it, and add more realism. I think you will find very similar result (that is, profit maximizing strategy will never be to compete on price). Except for the price wars, which (in my opinion) always happen only temporarily when the big company wants to drive smaller companies out of the market entirely to get more market share as a result (just like real wars). But you need to have inelastic demand for that, otherwise I doubt it can be considered a viable strategy.
I came across Keen awhile ago, and actually made my own model of the price setting algorithm in python and confirmed his results. My beef is that it is an unrealistic constraint that the same price setting algorithm will be used by all companies. And it is possible for someone to enter the market using a different algorithm.
MR!=MC in the real world, but the classic micoecon explanation for this is monopolistic competition. I'm not sure why there is a need to say that all of economics is wrong. Are there commodities markets with open entry and no open collusion where MR!=MC? (Where marginal cost is defined as the cost of bringing on a new production line, not just producing one more unit).
Non-commodity markets get quite complicated. Often the market bifurcates into a set of premium brands and a set of low cost brands. Premium brands often try to get a monopoly over one subset of customer, who need some special feature, or try to build their brand as something special, or otherwise differentiate. Premium brands explicitly do not compete on price. Salesmen will say stuff like, "Look I know we are pricy. If you want the el cheapo software, don't buy us. Buy us if you want the good stuff."
If the market just has premium brands, that all price way above cost of production, someone is going to try to enter the market with a lower cost product.
I never gathered what point Keen was trying to make, not what he was saying that was both novel and true.
> And it is possible for someone to enter the market using a different algorithm.
By the way, I think this often happens in the real world, where some person irrationally (from the economist's point of view) provides a valuable, yet underpriced (with respect to the real demand curve) service, because he is just being misinformed, idealistic or is interested in success for its own sake, not for the sake of money. Or it's just that the product is new and actual demand curve hasn't been established yet.
Absurdly, this is then hailed as an example of market efficiency, while there is nothing rational in such behavior from the game theory perspective (unless you consider much wider frame of how human morality evolved etc.).
I think Keen makes two points. One is (as it often is the case in Debunking Economics) that this is "surprising" for students of economics. He cites various textbooks, like Mankiw. If economists know these things, why do they lie to students? It is indoctrination, plain and simple.
The second point, IIRC, is that the Cournot model of perfect competition is mathematically wrong. The assumptions that no firm can influence the price and decreasing demand curve are contradictory, regardless whether or not these provide a good model for the real world.
Or, in this case, to play, but play a safer game, where algorithms find their a local minima, and profits are potentially smaller than when using an aggressive stance, but maximized in the long run.
What is the penalty for defecting from the group to clear the market? An actor gets kicked out of a group in which he was a voluntary member?
Suppliers take a risk by price-fixing in that they may be underutilized, or be left with excess supply. Breaking even occurs when a few customers paying the collusion price net the same income as many customers paying a spot price. Clever suppliers can profit, but all actors are still independent. The only recourse for clearing the market is being left out of future collusion, which means nothing if you're clearing 100% of your supply and the adversary is clearing 0. The only choice of all actors is to compete for satisfaction.