[ddinh]

In terms of reducing the ability of speedy traders to gain an advantage in a continuous time market, how does IEX's fixed time delay compare to discretizing the timesteps, so all the trades within some finite interval [t, t+e] are treated as if they came at the same time?

This talk presents compares discretization to standard continuous-time bidding, but doesn't go into a lot of detail about how it compares to IEX-style delays: https://simons.berkeley.edu/talks/eric-budish-2015-11-19

[chollida1]

The IEX delay and batch auctions are really two completely different things.

IEX delays almost all messages in and out by a small amount.

They don't delay messages for orders that their own router modifies, their discretionary peg order. This lets their peg orders update before any one, ie HFTs, can update their own resting orders on IEX or at other exchanges in the case of a fill at IEX.

It's important to note that even with the IEX design, speed is still very important as its still a price time priority queue for order placement and cancelling. An analogy would be if all 100 meter sprinters reaction times were delayed by 1 second from hearing the starters pistol. The fastest runner still wins, its just that all reactions times are delayed evenly.

Batch auctions are a whole different animal, and to be honest, one I'm not very familiar with. Their greatest weakness is that they don't operate in a stand alone environment, ie the rest of the world continues to trade around them which eliminates most of their advantage.

[sjbase]

Do you mind expanding on that last point? I'm interested in your take on how/why a batch auction's advantage is eliminated by the rest of the world trading around them.

[kasey_junk]

Largely because the entirety of latency arb has to do with taking advantage of price differences between exchanges. How those differences are created is largely immaterial.

As long as you have multiple exchanges that are not perfectly syncd (a distributed systems problem!) There will be time based arbitrage opportunities.

[rrrrsss99]

Not exactly in all products. US equities exchanges have to comply with the nbbo (national best bid offer), so they're required to forward an order to the exchange with the best price. This isn't true for futures, but there aren't any fungible futures in the US. This means that latency arb in US equities isn't true arbitrage. If you see a price level get sweeped at one exchange, you may guess that the others will follow, but it's no guarantee and most sophisticated traders who trade the volume to do this will trade in a dark pool or block trade, so the opportunities are close to non existent. The intent for equities market making is to trade retail orders with the assumption that they they're placed with no knowledge of a true price and won't move the market. The true arb opportunities today are reserved to minis vs futures or cross exchange futures vs indexes

[tptacek]

I think the point Kasey is hinting at is that the functionality you claim RegNMS requires exchanges to implement is not, due to the CAP theorem, actually possible to do reliably.

[harryh]

"Pro-rata to break ties" <--- A pretty big issue that he quickly glosses over.