4 ms·

Adversarial competition and collusion in algorithmic markets

48 points by theptip 3y ago

AtlasBarfed 3y ago
I can imagine that with the rise of AI, you now have the basis of plausible deniability for insider trading: "The AI told me to do it" I think this will be an interesting legal case when it (inevitably) happens. Because AI is almost impossible to introspect/poll from the outside. If you query it, that is a different "neuron path" and it can't be relied upon to know what/how/why it "knows what it knows". It can also be plausibly denied that the CURRENT AI response doesn't say what you did. Just say it was retrained / modified to any degree (and of course you don't keep backups), and even if the AI doesn't produce the same answer, you can simply say the training has changed or the inputs have changed (you also don't back up the historical input values).
- tommiegannert 3y ago
  Legally, I'm guessing this is the same defence as "I just threw darts on the board and got lucky." You still decided to trade an instrument you had insider knowledge about. You'd better have a paper trail to show that the dart wasn't biased.
- gruez 3y ago
  >I can imagine that with the rise of AI, you now have the basis of plausible deniability for insider trading: >"The AI told me to do it" Do you even need to use that as a defense? Doesn't the SEC have to provide affirmative proof that you traded on material nonpublic information? Have they ever convicted someone of insider trading on good trading records alone? Otherwise, I don't really see how using an AI to launder stuff makes sense. You need to acquire that material nonpublic information somehow. Because it's nonpublic, you're probably going to have to go out of your way to get a hold of it, so that's how the SEC is going to nail you. Feeding a bunch of leaked quarterly earning reports into an AI isn't going to let you off the hook.
- cma 3y ago
  It's a bigger deal for wash trading which relies on intent of each trade.
janalsncm 3y ago
As a concrete example, Uber and Lyft can likely drive up prices without any price fixing or collusion simply by understanding the maximum price customers will pay and pushing it higher. Simple explore/exploit.
- FrustratedMonky 3y ago
  IT is happening in rents. A lot of rental properties have started using a tool to determine what the 'market can bear'. And have started raising rent. Their is no collusion. But they are all using the same technology, which is telling them to raise rent. Since they are all raising rent, everyone is trapped and must pay, thus the 'market' can bear the price, since the software can tell everyone else is increasing rent, then it must be possible. Kind of self fulfilling prophecy.
  - Enginerrrd 3y ago
    I don't think that would work. The Uber/Lyft thing works because there is such a thing as a pseudo-monopoly pricing strategy that is optimal. This doesn't work in markets with a large number of actors. Now... MAYBE it could work if you have a very large fraction of rentals consolidated under a small number of listing services?
    - FrustratedMonky 3y ago
      I'm not sure the percentage of the market owned that is needed to influence the total market. I'm sure different markets have different thresholds. I think you are thinking that all renters are 'actors', so there is to many individual choices, thus this can't happen. But, here the actors are the rental property owners. That is a much smaller number. IF you only have a few property owners, and they all use software telling them to increase prices, then the total market goes up, and individual renters do not have any options to go anywhere else, so must pay. It really is the same problem as with collusion, or price setting, it is just here, what is happening is 'the colluders' all happened to be using the same software. They aren't coordinating, they are coincidentally all being pushed in same direction by a single software company. https://www.propublica.org/article/yieldstar-rent-increase-realpage-rent https://www.propublica.org/article/yieldstar-rent-increase-r... >" RealPage became the nation’s dominant provider of such rent-setting software after federal regulators approved a controversial merger in 2017, a ProPublica investigation found, greatly expanding the company’s influence over apartment prices. The move helped the Texas-based company push the client base for its array of real estate tech services past 31,700 customers. The impact is stark in some markets. In one neighborhood in Seattle, ProPublica found, 70% of apartments were overseen by just 10 property managers, every single one of which used pricing software sold by RealPage."

48 points by theptip 3y ago

I can imagine that with the rise of AI, you now have the basis of plausible deniability for insider trading: "The AI told me to do it" I think this will be an interesting legal case when it (inevitably) happens. Because AI is almost impossible to introspect/poll from the outside. If you query it, that is a different "neuron path" and it can't be relied upon to know what/how/why it "knows what it knows". It can also be plausibly denied that the CURRENT AI response doesn't say what you did. Just say it was retrained / modified to any degree (and of course you don't keep backups), and even if the AI doesn't produce the same answer, you can simply say the training has changed or the inputs have changed (you also don't back up the historical input values).

tommiegannert 3y ago
Legally, I'm guessing this is the same defence as "I just threw darts on the board and got lucky." You still decided to trade an instrument you had insider knowledge about. You'd better have a paper trail to show that the dart wasn't biased.
gruez 3y ago
>I can imagine that with the rise of AI, you now have the basis of plausible deniability for insider trading: >"The AI told me to do it" Do you even need to use that as a defense? Doesn't the SEC have to provide affirmative proof that you traded on material nonpublic information? Have they ever convicted someone of insider trading on good trading records alone? Otherwise, I don't really see how using an AI to launder stuff makes sense. You need to acquire that material nonpublic information somehow. Because it's nonpublic, you're probably going to have to go out of your way to get a hold of it, so that's how the SEC is going to nail you. Feeding a bunch of leaked quarterly earning reports into an AI isn't going to let you off the hook.
cma 3y ago
It's a bigger deal for wash trading which relies on intent of each trade.

janalsncm 3y ago

As a concrete example, Uber and Lyft can likely drive up prices without any price fixing or collusion simply by understanding the maximum price customers will pay and pushing it higher. Simple explore/exploit.

FrustratedMonky 3y ago
IT is happening in rents. A lot of rental properties have started using a tool to determine what the 'market can bear'. And have started raising rent. Their is no collusion. But they are all using the same technology, which is telling them to raise rent. Since they are all raising rent, everyone is trapped and must pay, thus the 'market' can bear the price, since the software can tell everyone else is increasing rent, then it must be possible. Kind of self fulfilling prophecy.
- Enginerrrd 3y ago
  I don't think that would work. The Uber/Lyft thing works because there is such a thing as a pseudo-monopoly pricing strategy that is optimal. This doesn't work in markets with a large number of actors. Now... MAYBE it could work if you have a very large fraction of rentals consolidated under a small number of listing services?
  - FrustratedMonky 3y ago
    I'm not sure the percentage of the market owned that is needed to influence the total market. I'm sure different markets have different thresholds. I think you are thinking that all renters are 'actors', so there is to many individual choices, thus this can't happen. But, here the actors are the rental property owners. That is a much smaller number. IF you only have a few property owners, and they all use software telling them to increase prices, then the total market goes up, and individual renters do not have any options to go anywhere else, so must pay. It really is the same problem as with collusion, or price setting, it is just here, what is happening is 'the colluders' all happened to be using the same software. They aren't coordinating, they are coincidentally all being pushed in same direction by a single software company. https://www.propublica.org/article/yieldstar-rent-increase-realpage-rent https://www.propublica.org/article/yieldstar-rent-increase-r... >" RealPage became the nation’s dominant provider of such rent-setting software after federal regulators approved a controversial merger in 2017, a ProPublica investigation found, greatly expanding the company’s influence over apartment prices. The move helped the Texas-based company push the client base for its array of real estate tech services past 31,700 customers. The impact is stark in some markets. In one neighborhood in Seattle, ProPublica found, 70% of apartments were overseen by just 10 property managers, every single one of which used pricing software sold by RealPage."