Comment on the July 27 prediction: Regarding sliding prices, for the most part CORRECT, although there have been scattered hikes around the Midwest.
Sunday, August 11, 2024, 7 AM: It looks like wholesale prices have dropped 50 cents a gallon the past two weeks, as the refinery in Joliet has come back online. Average retail prices have been working their way slowly lower, and I would like to see that accelerate this week. We are down to $3.29 in Greenville, MI (which is near Grand Rapids) and I’d like to see many more stations adopt that price.
But, I can’t make that prediction today, because of the higher-margin world we have been living in this year. That brings me to my last post on July 27, I wondered whether some sort of artificial intelligence program is now being used to set gas prices — collusion via algorithm. I had no direct evidence for this idea and I was just trying to suggest a reason why playing the Gas Game has seemed different this year. Not surprisingly, I am not the first person to think about this question when it comes to gas prices, and in fact, a research article came my way this weekend that looks at algorithmic pricing (AP) of gasoline in Germany, starting in 2017. Here is an excerpt that I think gives away the story:
“Simulation results … suggest that it can take a long time for algorithms to train and converge to stable strategies. During this time, algorithms may learn to punish competitors for reducing prices or other tacitly-collusive strategies.
We find evidence consistent with this. Margins do not start to increase until about a year after market-wide adoption, suggesting that algorithms in this market learn tacitly-collusive strategies. We also examine the pricing behaviour that emerges in markets where both duopolists are algorithmic adopters. We show that in a market where both duopolists adopt, a station is more likely to respond to a rival’s price decrease with an immediate price decrease of their own. There is no comparable change in the propensity to respond to price increases by a rival.
We also find that when both stations adopt algorithmic pricing, the duopolist setting higher prices is less likely to undercut the duopolist setting lower prices. The timing of these effects is consistent with the timing of the price and margin increases. Altogether, these findings provide further evidence that adoption affects competition and they suggest that the algorithms learn that undercutting will not be profitable, since the undercutter will always be followed to the lower price by its rival.” [bold is mine]
Duopoly … two major players in the market. Hmmm, sounds familiar. If we are dealing with an AP-powered duopoly in the Midwest, and these results from Germany are transferable, that would suggest from my data that there was some training of algorithms in 2023, leading to larger margins in 2024.
Again, this is all speculative, but I would not be suprised if some economist is working on reproducing the German study with the markets I follow.
Readers may also be interested in this new article in The Atlantic on the subject of AP. -EA