HFT Myths

[hft]

Do you use VPIN to detect short-term imbalance, say within 30 seconds? Could you elaborate more on the edge conditions which are not listed in the VPIN wiki article?

BTW, do you have any experience on the high order-to-trade market making strategies in case you get penalty on issuing a large number of cancels?

I don't use 'VPIN' strictly as it's defined in some articles, but yes I use Last Trade information from the exchange to formulate a short-term imbalance signal, in the neighborhood of 30 seconds. An edge condition might be a huge single trade that comes through that skews your imbalance signal if it's using short/small windows. More generally, your signal needs to adjust to market conditions...your signal at 7 PM is going to behave differently at 7 PM than at 9 AM.

For handling messaging ratios, I just try my best to reduce flicker in my orders, at the expense of not always making an optimal market (wider or tighter than you'd like to be at times). I compare it to a gas station owner being too lazy walk out and adjust his price sign by a penny or so, at the expense of not having it priced exactly where he wants it.

 

[hft]

HFT

You can download the article (indicateing that VPIN doesn't predict near-term volatility) from the following link:
http://papers.ssrn.com/sol3/papers.cfm?abstract_id=1881731&rec=1&srcabs=2062450&alg=1&pos=3

I'd like to hear your comments.

I read the intro section. Makes sense. Their conclusion is:
"...adoption of any specific metric for order flow toxicity should be contingent on satisfactory performance relative to suitable benchmarks"

Duh.

VPIN is just a fancy name given to a certain method of aggregating trade information. There are a zillion ways to do it, and a good analyst can probably find some use for any of them, and vice versa.

I find it funny they can't publish the magical VPIN formula because the other guys patented it.

 

[xoxohth]

Simple example:
EdgeRequired = 0.1 ticks
MktImbalance = BidSize/(BidSize+AskSize)
TheoPrice = BidPrice + MktImbalance * (BidPrice + AskPrice)/2

MyBid = TheoPrice - EdgeRequired, rounded down to nearest tick
MyAsk = TheoPrice + Edge Required, rounded up to nearest tick

In practice, theoPrice contains signals from both the traded market (trades occurring in the market, etc) and other markets (SPY, etc).

This is an example of a signal based on trades:
http://en.wikipedia.org/wiki/VPIN

Do you care about Q spot when making the decision to place or cancel orders or does running such a tight spread around the mid price basically keep you joining early on new prices so you always have good Q position? Like does being 1st in ES have intrinsic value even if your signal says it will move against you?

How do you decide how much spread to ask for on something like ES? Clearly it must be < 1 tick since otherwise you'll flash orders constantly and never trade but how do you come up with the number? Guessing it changes when the market is whipping around?

Do you typically take liquidity to scratch trades and do you think this is a mean thing to do to all the manual traders out there?

Also how the hell can this strategy make money? You want to be bid when the bid is large? If that's your signal for sending an order how can you ever get trades except when some whale whacks it for 1000 contracts? Seems fake to me, no offense.

 

[hft]

Do you care about Q spot when making the decision to place or cancel orders or does running such a tight spread around the mid price basically keep you joining early on new prices so you always have good Q position? Like does being 1st in ES have intrinsic value even if your signal says it will move against you?

How do you decide how much spread to ask for on something like ES? Clearly it must be < 1 tick since otherwise you'll flash orders constantly and never trade but how do you come up with the number? Guessing it changes when the market is whipping around?

Do you typically take liquidity to scratch trades and do you think this is a mean thing to do to all the manual traders out there?

Also how the hell can this strategy make money? You want to be bid when the bid is large? If that's your signal for sending an order how can you ever get trades except when some whale whacks it for 1000 contracts? Seems fake to me, no offense.

1) Yes I care about Q Spot. Queue position is a signal in and of itself.
2) Arbitrarily pick a spread to ask for, adjust it up if you're getting picked off, down if you're not getting fills, and yes it adjusts with market conditions.
3) Many times I do take liquidity to scratch trades. No, I don't feel bad for manual traders. If you're not efficient enough to be resting orders to make the bid/ask spread then you should expect to get picked off (I certainly do often).
4) This strategy as stated does not make money. It's just an example of where to start with a basic, single-product market-making strategy. You have to add more signals and adjust accordingly.

 

[xoxohth]

4) This strategy as stated does not make money. It's just an example of where to start with a basic, single-product market-making strategy. You have to add more signals and adjust accordingly.

Ok so reading more of what you wrote you also have to look at other products to shift your prices around? I can see that working on something where you have other really correlated products to hedge with, or in the equities world where you can look at other markets, but still don't get how it can work scalping something on one market using its own book as your only "input" for pricing. You mention looking at things like trade rates. If something is trading bid aren't you by definition getting picked off at that time so how can you pull away? Seems like a tiny edge and you'll be wrong constantly but maybe I misunderstand how this game works?

Would you say most of your good trades are ones where you can look to something else to price? Like I get how being really fast would let you trade products with arbitrage relationships but seems a lot weaker if you're trying to pull away when everyone is trading, like playing high-stakes slapsies or something.

 

[rossw]

but still don't get how it can work scalping something on one market using its own book as your only "input" for pricing. You mention looking at things like trade rates. If something is trading bid aren't you by definition getting picked off at that time so how can you pull away? Seems like a tiny edge and you'll be wrong constantly but maybe I misunderstand how this game works?

Lots of different ways here. You could have good queue priority on both sides and make the spread (the ideal but rare). You could have bids stacked within the queue and when done on the front bids pull your back level ones before doing bad trades. You may even pull your bids and hit the remaining bid if you get enough of a signal (eg lots of trades on the bid, big offers, wind starts blowing from the West, whatever..) - that way other hft's might have a bad position, main seller follows it lower etc

Remember HFT has tiny fees so scratch trades are no problem. If he's getting rebates he may even make some money from doing scratches. And as already mentioned early, there are loads and loads of scratch trades within most strategies.

 

[maxima120]

Is there a way to estimate number of stop orders traded at a particular level.. from analyzing the CME tape? does anyone do that?

 

[stock777]

HFT SCUM admits that they cheat

http://www.zerohedge.com/news/2014-...irm-reveals-1-losing-trading-day-1238-days-tr

 

[hft]

Is there a way to estimate number of stop orders traded at a particular level.. from analyzing the CME tape? does anyone do that?

I don't know of a clear-cut way to do it. One possibility would be to look into situations where the market ticks towards a round number with strong size on it, and you see unexpected large number of trades that print at that size (i.e. mkt was 1.01 x 1.02, then downticks to 1000@1.01 x 10@1.02 but you see a bunch of prints at 1.01 anyway). Similarly you could come up with other scenarios that indicate stops being hit, but to really answer your question there's no flag on LastTrades from CME that indicate they are a result of Stop type orders.

 

[maxima120]

There's all sorts of ways to encrypt/permission files. .

what about physical security? how do you protect your servers at colocations? Do you rent whole cages? Do you have special legal agreement with the colo provider/exchange? Have you ever heard of an actual cold boot attack on someones server (within the HFT community or close)?

Thank you