High-Speed Trading: Profit—and Danger—in Milliseconds

Eric Scott Hunsader has gone completely down the rabbit hole, and he doesn’t like what he’s finding there.

Hunsader is the CEO of a Chicago-based market analytics firm that specializes in high-frequency trading—super fast trades executed at the speed of light that can alter asset prices faster than human beings can react to the changes.

Based on his own analysis, Hunsader has come to a startling conclusion: Markets today are even more susceptible to sudden failure than they were two years ago during the “flash crash,” which brought the stock market down by about 1,000 points in mere minutes.

That’s because a new breed of trader armed with hundreds of millions of dollars to deploy is trading so fast—and with such spikes in volume—that he can dry up liquidity in an instant, causing severe price swings.

To explain to a lay person, Hunsader offers up two examples of the kinds of trades he’s seeing. The first happened less than a minute before the April Jobs report was released by the Department of Labor in Washington.

That report is traditionally one of the most dramatic market-moving events of each month. As a result, traders tend to lie low in the minute or so before the number comes out at 8:30 a.m. on the first Friday of each month, so they don’t get caught when the market changes.

But Hunsader argues that he’s seeing a small group of high speed traders who aren’t lying low. In fact, they’re taking advantage of the regular and predictable lull in the market to pop high speed trades in order to intentionally create a several hundred millisecond burst of volatility, and then execute follow-on trades to profit from that.

To understand what happens, you have to go inside just one second of trading and look at the way markets move at speeds that can be almost imperceptible to human beings.

On May 4, Hunsader , he spotted those traders just before the April number was released. At 8:29:20 and about 200 milliseconds, he says, someone — he has no way of knowing who — executed a trade in the five year T-note futures market worth about $150 million.

A chart of that single second in the market shows that prices are relatively stable until the trade. And just after that, for the rest of the second, prices spike, and gyrate up and down as other automated high speed computers react to the trade.

Hunsader says he doesn’t know exactly how the traders make money off the volatility that they create, but he suspects they’re making other trades in the milliseconds following their market moving trade that take advantage of the relationships between this market and others that are impacted by it.

The traders that move first, and fastest, win, he says.

“It’s like two guys running in the woods, and they see a bear and one guy drops down and puts his shoes on and the other guy says, ‘what are you doing that for, you can’t outrun a bear,’” Hunsader says. “And the guy goes, ‘I don’t have to outrun the bear, I just have to outrun you.’”

On another occasion, Hunsader says he saw traders taking advantage of something as fundamental as the speed of light.

Because trades are executed by fiber-optic cable, the fastest they can travel—and the fastest anything in the universe can travel—is the speed of light. But even at that speed, it takes about 11 milliseconds for information from exchanges based in New York to get to exchanges based in Chicago. And that provides an opportunity for arbitrage for those who can move fast enough.

http://www.cnbc.com/id/47432428

 

I don't get it... why they just don't stick to simple stuff that REALLY works -like moving averages, divergence, trend lines , indiacators like MACD,RSI and all that..
Somehow they seem to disregard classic AT and it is not because it's DEAD 'cos it is not... or perhaps it is???

 

and this is exactly why articles like this to me are a bunch of baloney.

What good is telling me someone executed a relatively large order before a report came out? I mean, who gives a crap. If they are gaming anything, they will be gamed right back by other HFT's and so they'll get burned.

I swear.

 

The article goes on to say...

Umm, and? 1300 ES contracts is like trading 100 shares of IBM in terms of liquidity.

See above

Im not sure what he's saying here? Is he trying to say that some trader can predict the direction of the market (predictable fallout?) from a jobs report and is making a fast bet ahead of everyone? I don't think so and so what anyway. And if its some kind of ES/SPY arb strategy.. so what? Makes the market more efficient.

I really dont get it. I'm looking for that part about how the market is going to go completely bonkers and I cant find where he makes a case for that other than to say, the most liquid securities in the world, treasuries and s&p futures/etfs are being traded really fast.

Is that all you got?

 

I agree and hope their "market analytics" is of a higher caliber when it is being presented to paying clients.

 

What he is trying to say is that if you buy 1300 ES contracts, it takes 11 milliseconds for that information to reach New York, which is ample time for you to prepare for the jump in prices that the SPY is about to experience due to other HFT arb-algos thinking that's what they need to do. The buying of the ES contracts thus pays off (hopefully) since you are able to predict a move in SPY, 11 milliseconds before it happens.

Get it?

 

No, not really. (hopefully) is not a strategy. And it still doesn't tell me where the market is going to go haywire.

 

The problem with that assumption (that the SPY will experience a jump in prices) is that in the 11 milliseconds that it takes for the info to reach New York, somebody else can sell 2600 ES contracts, meaning that by the time the info about the 1300 ES contracts reaches New York, it's old.

Alternatively, anyone with an interest in the relative prices of the ES and the SPY will locate the team responsible for profiting from those relative prices in Chicago. Those who keep their teams in New York will either relocate or go out of business from their losses.

There is just so much excuse-making and fear-mongering in this business it's ridiculous. Not you, per se, but just generally.

 

it's simple you buy 1000 es at 1347 and you have a market offer for 1000 waitng to enter,the traders buy thinking it was the emp report and you make a quick point or more before the report comes out,if your initial order was in line with the report, more ,it's just a quick way to make a buck using the speed of your order entry ,the haywire part is if the other or several hft programs are setup to buy or sell when a 1000 lot hits the bid/offer,if so you win and screw them up,now if there is another rule that kicks into affect when a 5000 lot hits,and all of the sudden a chain reaction happens, you have may 6th deja vu...the guy putting in the initial 1000 lot is messing with the laws of the hft programs universe..since we can't know how the program is setup, it's only speculation

 

That article is so riddled with errors it is only going to be understood by someone that already understands what is going on.

You got half the strategy though. The delay works both ways, so the buy SPY trade will also be seen by the Chicago boxes and act on it on the ES, AND the buy ES trade will be seen by NY boxes and act on it on the SPY. So at both ends the trade was executed at the same time, and immediately on confirm from the matching engine an order was put in to sell at some incremental price higher on the same matching engine. The SPY sell trade would be taken out by the dumb boxes in NY that see the ES trade 11 milliseconds later, and the ES sell trade would be taken out by the dumb boxes in Chicago that see the SPY trade 11 milliseconds later. Even if you scratch or lose a little on one side, you only need to catch the other side for a handle or two.

In fact, you don't need (super) fast connections to do initiate this trade, just lots of money to paint the tape and idiot programs that can't tell inside information from market manipulation. Another possibility is taking advantage of clock synchronization errors, which WOULD be clever, since [poorly implemented] machines would believe the report came out, when in fact it hadn't.