The Game Theory of How Algorithms Can Drive Up Prices (quantamagazine.org) 16
Computer scientists at the University of Pennsylvania have proved that pricing algorithms can drive up prices even when they lack the capacity to collude. Aaron Roth and four colleagues studied so-called no-swap-regret algorithms, which are designed to minimize losses and were previously thought to guarantee competitive pricing. The researchers found that when such an algorithm faces an opponent using a nonresponsive strategy -- one that randomly selects from predetermined price probabilities without reacting to competitor moves -- both players can end up in equilibrium at high prices.
Neither has an incentive to switch strategies because their profits are nearly equal and as high as possible under the circumstances. The nonresponsive strategy cannot express threats because it does not respond to opponent behavior, yet it effectively coaxes the learning algorithm into raising prices. Mallesh Pai, an economist at Rice University not involved in the research, said the finding matters because regulators have no clear grounds to intervene without evidence of threats or agreements. Roth conceded however that he lacks a solution to the regulatory challenge his team identified.
Neither has an incentive to switch strategies because their profits are nearly equal and as high as possible under the circumstances. The nonresponsive strategy cannot express threats because it does not respond to opponent behavior, yet it effectively coaxes the learning algorithm into raising prices. Mallesh Pai, an economist at Rice University not involved in the research, said the finding matters because regulators have no clear grounds to intervene without evidence of threats or agreements. Roth conceded however that he lacks a solution to the regulatory challenge his team identified.
Didn't See That Coming! (Score:4, Interesting)
At least we have evidence and not just logic based on a couple hundred years of capitalism and thousands of years of human greed.
Re:Didn't See That Coming! (Score:4, Insightful)
Sure, if by "evidence" you mean simulations of a two-seller market where one of the two sellers in a market has a strategy that is essentially only flipping a coin.
This is evidence about the sorry state of economics research much more than about the real world.
Re: (Score:2)
I didn't realize the study was quite as thin as you point out, but my point still stands.
Re: Didn't See That Coming! (Score:2)
Can you do representative agent models in standard economics next?
It works the other way too (Score:1)
real world??? (Score:3, Interesting)
Re: real world??? (Score:2)
How excited do you get about the toy Econ 101 models that prove prices are just rational signals of supply and demand conditions, which models the Fed uses to set interest rates etc.?
Re: (Score:3)
So, Boeing and Airbus have huge numbers of competitors. And the US media, oh, all of four companies.
Re: (Score:2)
Re: (Score:2)
"Almost nothing in the real world is a 2-player game"
That's quite untrue if you ignore rounding-error players in a substantial number of markets.
Yes, real world (Score:2)
You can also read ProPublica's analysis [propublica.org].
But you seem to be more interested in attacking theory, so let me ask you this: If algorithmic pricing doesn't increase prices, why would landlords and retailers use it?
Re: (Score:2)
Re: (Score:1)
Re: (Score:2)
Tax policy (Score:2)
Tax policy will need to index prices to inflation and raise rates progressively as one deviates from the inflation rate. In the case of things such as rents, one will have to tax vacancies at increasingly punitive rates.
It could be regulated (Score:2)
I mean, we regulate several other algorithms (e.g. accounting is just a regulated algorithm).
We could make them be open-source and enforce certain properties, there's a lot that could be done to mitigate the effects or even ban them outright.