[mschuster91]

Doesn't surprise me the least bit. Where there is something available to exploit (in this case, access to direct, fast feeds), it will be exploited.

Would it be possible, legally and technically, to put a special additional fee/tax onto high-frequency trading while leaving normal high-volume traders alone?

[mrgordon]

Its certainly possible technically (and legally in some cases). Italy has a 0.02% tax on transactions lasting less than half a second: http://www.dw.de/italy-first-to-slap-tax-on-high-speed-stock...

[noonespecial]

Go to the router just entering the exchange, type:

  tc qdisc change dev eth0 root netem delay 100ms 10ms

Problem solved.

[DevX101]

what does this do? (non-finance guy here)

[jackvalentine]

The command introduces a delay of between 100 and 10ms to every packet. Screwing up trades that rely on being the fastest.

http://www.linuxfoundation.org/collaborate/workgroups/networ...

[noonespecial]

Its a bit of a joke. Its a network thing, not a finance thing. What it does is adds a delay and more importantly, a bit more random delay to the time it takes the order to reach the exchanges server. Once you add in the non-determinism, HFT basically falls apart because you can't take your truckload of cash and buy yourself a place in a datacenter that's 2ms closer to the exchange and front-run everyone.

It would be a wonderful thing to see all of those millions these guys have "invested" shaving a millisecond or two off their transaction times laid waste by a single command.

[dzderic]

This "solution" will only make it harder for regular folks to execute orders, since HFTs will beat the randomness by shooting multiple orders through multiple order gateways.

[noonespecial]

Guess we'll need a hierarchical token bucket with stochastic fairness queueing as well.

We don't just need it to be random. We need there to be no way of ever quite knowing if any given order will beat another order to the exchange (within a given time period, of course). They won't know if they can beat joe ordinary, and they definitely won't know if they can beat the other HFT's. That might be enough to put a lid on it.

Edit: For those playing along, here's the metaphor. Joe goes to market to buy sheep. Bill knows Joe is going so he sends a fast runner ahead of him to buy the cheapest sheep in town first so he can mark them up and sell them to Joe when he arrives. We try making everyone wait at the town gate for a random amount of time to give Joe a chance to arrive and get through. So Bill (being very rich) just sends 10 guys so one is very likely to be let in before Joe anyway. Next we introduce the stochastic filter. We make everyone line up and then shuffle the order every once in a while, but Bill still has more guys so he might still get one in first more often than not. Finally, we add the token bucket. For every one guy that we know employed by Bill admitted, we make the next one wait twice as long to get in, so if Joe and 10 Bills show up, Joe and the first Bill are essentially on even footing again because the 2nd through 10th Bill would have to wait too long to matter.

[gd1]

"For those playing along, here's the metaphor. Joe goes to market to buy sheep. Bill knows Joe is going so he sends a fast runner ahead of him to buy the... "

Seriously, how many times have we discussed this issue on this site and we still get this bullshit. Bill doesn't know Joe is going. He doesn't. Get it through your thick heads.

[simplemath]

Pretty sure he's suggesting adding a non-determistic lower bound to trade frequencies at network boundary

[harryh]

Not really no.

High Frequency Traders are selling a service (liquidity). A tax on them is mostly just a tax on their customers.