Hacker News Reader

Comment by joe-mccann | Hacker News Reader

Welcome to the world of High Frequency/Algorithmic trading.

In this case, essentially every person that had a "stop loss" order was just hunted down by a wave of HFT programs and all those stop loss orders were triggered, which send "market orders" which means, fill my order at the current market price.

When everyone does this at the same time, there are more sellers than buyers so prices drop dramatically. However, this presents an opportunity for HFT programs especially when they have similar strategies, meaning they all are doing the same thing, pushing the market in the same direction. Now, the HFT programs forced people to get stopped out, then they bid the market up and buy, which pushes it right back to the prior level.

Take a look at AAPL, RIMM, GOOG, SPY, etc. and you'll see it is all the same pattern.

View on HN

I know people in the financial industry, and a HFT "blip" is not the cause of this.

So what is it then?

per mschwar99 below

CNBC: Trading Error at Major Firm Blamed for Market Plunge: Sources (Story Developing)

Honestly, if you don't know that CNBC is to the financial markets as what the Hee Haw is to the news, then I can't help you.

CNBC does NOT know what the cause of this was. The exchanges will come out with some "computer glitch" story or the "fat finger" excuse, but neither of those are true. FWIW, the fat finger excuse is, in a nutshell, some trader accidentally pressing the wrong button on the trading desk.

The market was at a technical support level with various moving averages, timelines, etc. all coming into play. Many people had/have stop losses because the market was/is looking due for a correction. Algo/HFT pounced on this, collectively, and thus the massive drop.

Trust me, if you are buying into what CNBC tells you, you are really missing the truth...

"A possible culprit for the drop was a trader error in which someone entered a "b" for billion instead of an "m" for million in a trade. Multiple sources confirmed the report to CNBC and CNBC.com"

http://www.cnbc.com/id/36998463

That sounds hard for me to believe, but thats one rumor. That wouldn't account for the 10% drop but it might have put initiated the drop and then caused other people (and computers) to panic and sell.

Considering that the volume on Proctor & Gamble trading today not only does not include a spike of a billion shares sold but does not look different from relatively recent trading days, I'd chalk this one up to just a rumor and nothing else.

Today's chart does show a spike down to $50 or so, maybe even lower. Go to finance.google.com and look at the intraday chart. It also shows a volume spike, but curiously that's later.

[Edit] Actually the stock was down at $39.37 from $61. On normal days PG barely moves at all.

There's undoubtedly a price spike. But to confirm or deny the s/million/billion/ explanation we are looking for a volume spike at the same time.

Don't forget all the derivatives linked to the price. You don't see that volume or the losses caused by that.

It depend on the resolution of the chart. If the resolution is minute-based, then it could have hit $0 then come back up to $39 within a minute. You wouldn't know. You'd have to have a special high-resolution stock feed.

Yes, but I would expect that the low and high for the day shown on the finance websites does show everything down to tick level. Not real time of course.

"That wouldn't account for the 10% drop"

Welcome to the world of positive feedback loops, enjoy your stay.

Doesn't that kind of software have annoying dialog boxes saying "Are you sure you want cause financial meltdown?"? Or at least a less intrusive "Did you mean 'm'?"?

It probably should.

I think the first one is actually the better choice.

It's really surprising, assuming this is really the cause, that a single letter typo could have such an effect and this has 1) never happened before and 2) not been caught in testing.

More than likely it is a cover story they hope will catch on enough to either cover the panic, or the real reason.

I will tell you what though - someone... someone, made a lot of money from this. Inadvertently, but more than likely otherwise.

I doubt it's a person that made money. Usually individual investors lose big time when stop loss orders are activated, it's the institutional investors that win.

Stuff like this does happen now and again. One example is putting the wrong value in the wrong fields -

http://news.bbc.co.uk/2/hi/business/4512962.stm

In the above case the trader got the price and number of share fields confused.

No.

I have actually worked on such software. When a price comes in - a quote - you have a fixed amount of time to respond before it goes stale, usually a few seconds. It's assumed you know what you're doing and that you want to move quickly. Same as operating any other high-powered machinery.

I remember a feature request from a client of ours, we had keyboard shortcuts of k for thousand, m for million and b for billion. They wanted t as well.

Rumor was that someone at citi fat fingered 16 billion of eminis instead of millions. Even if it were true, it would not cause as much as a dent in it. Market was on a sliding selloff already. My suspicion was that this was a manipulation or HFT gone wild or both. If there weren't that spike down, dow would be at -5% easily, but what happened was -9 and then up to -3.

That'd be some mighty fat finger to hit the b and not the m... without hitting the n.

> That sounds hard for me to believe, but thats one rumor.

Would this rumor create insane profits for some? If that is so, those getting the profits should be investigated. Some employee could have been promissed a large sum in a off-shore account to enter a 'b' instead of an 'm'. It should look like an 'oops' to everyone else while a bunch of HFT algorithms could rake in serious profits as everyone's stop-loss triggers get hit. The employee might get fired, but he the pay-off was large enough, he might not need to work ever again.

mille mille vs. mille mille mille? http://news.ycombinator.com/item?id=1321405

Don't they enter the number itself, i.e. 1000000 instead of 1000000000. Doesn't seem so easy to get wrong.

The Dow Jones is made up of only 30 stocks. A $20 move in P&G would have a large instantaneous impact on the index which could then trigger the algorithms to go nuts.

DON'T PANIC

Of course HFT is not the root cause, it might have burned the fuel(stops) when acceleration approached terminal speed. However last tuesday and even before that VIX was on record heights and it pushed even higher today. No fat fingers then. Market is trying to tell us something, obviously. Cumulation was there, this was just a first in a series of explosions triggered by one reason or the other (greece/eu/trichet inaction today).

I never said anything about a "blip" and it is far from it. The equities markets are easy to manipulate; all you need is more $$ than the other guy and if all your friends are on board, that's a lot of $$ to move the market in the direction of your favor.

One of the reasons the forex markets are MUCH more difficult to manipulate is that you have $3.5 trillion traded daily. Not sure who can simply move that market. Equities on the other hand, much easier....

So, wouldn't this constitute a major transfer of wealth from individual investors with stop-loss orders to institutional investors like Goldman Sachs et al whose computers detected the dip and bought the stock back up?

From a lofty level, yes.

Just another day on Wall Street.

I knew Gordon Gecko had something to do with this.

A similar incident happened in Tokyo Stock Exchange a few years ago.

(...) on December 8, 2005 in which a Mizuho dealer mistakenly sold 610,000 shares in J-Com, a recruitment business, at 1 yen instead of selling 1 share at 610,000 yen.

The dealer, understood to be a 24-year old woman, immediately realised the mistake but was unable to cancel the trade because of a technical glitch in the exchange's trading system. Mizuho was forced to buy back the shares, leaving it with a 40.7 billion yen loss, prompting the firm to cancel year-end bonuses for all staff in the trading unit.

Note in Japan executed trades cannot be canceled (though in this case the trader was supposed to be able to cancel the orders before they were executed). Many brokers made a big profit. Some institutional investors too. One particular day trader became quite rich and to this day is a sort of legend on internet trading boards.

Er, I meant some "retail investors too".

CNBC now reporting on air that a human error by a trader at a major firm triggered the interday low.

Breaking News of cnbc.com reads:

Trading Error at Major Firm Blamed for Market Plunge: Sources (Story Developing)

The rumor is that the Citi e-mini desk sold $16B of futures ($SPY) by mistake and meant to do $16M.

(I don't buy it. CNBC is propaganda.)

I don't see how the order could even be placed. An emini contract has a 50x multiplier for SPY. That's about 50,000 per contract; that's a lot of contracts. I don't think such an order could be entered without error at Citi's desk and it would damn sure never pass checks at the exchange. Even if it was not flagged as odd, they would still be required to have on hand around 1.6 billion in collateral at very high leverage.

Of course, a brokerage like Citi will run most of their trades larger than 1000 lots through their program trading, which splits the order into lots of smaller trades of 100-200. I agree that it sure sounds suspicious, but I seriously doubt the exchanges saw a single bid for 1.6b shares of the e-mini.

Plus what system makes you type in "billion" or "million" instead of "1000000000" or "1000000"?

I call B$ on this.

I call B$ on M$

any decent trading system. When millsecs mean the difference between millions of dollars, having a trader type in all those zeros, then count them, would be costly time.

Fast traders don't click on icons and enter values, they enter an entire trade on one line.

Buy 1000 shares of Apple computer at 250.05 might be entered as:

buy1000aapl250.05

fair point, not in finance so didn't consider this.

However I find it extremely foolhardy that mistyping one character has an impact one thousand times greater, and that such an action can be performed without some kind of validation (whether explicitly via a confirmation, or implicitly in how the UX is designed).

Then again it is probably because of the typo risk that bankers get paid so much ...

`rm -rf` vs `rm -ri` could have that sort of impact. Or `sleep 1; shutdown -r now` vs `sleep 1h; shutdown -r now` to reboot the production server.

I once deleted all my source code instead of backing it up to my floppy by doing "del ." while command.com was in wrong drive. (Should have done "del a:\."; two letters cost me a night of running an undelete program)

I like my Bash tools as much as the next HNer, but always think it's faster, safer, and more accurate to use a windowed gui for moving or deleting important files and folders.

So what you are saying is they type it in with zeros.

Maybe you should have had an example not contrary to what you said.

That sort of command line looks like a recipe for disaster.

You should see the command lines the Navy uses for targetting (I'm not kidding).

I don't think we are doing millisecond-resolution trades by hand.

I worked as a trader, and I can confirm that this is not the way trades are hand-entered.

....on your platform.

...on my platforms.

Text based systems: FBSI, SIS

GUI based systems (when not pointing and clicking): Patsystems JTrader<A>, Trading Technologies' XTrader<A>

<A> These are systems I have used for personal use but not at work

I agree with this. Each platform can be completely different. Banks often have their own internal platforms too. Some are even set to trade in millions by default. So when you enter an order for 15 shares, you're really buy 15 million shares and so on.

A little off topic, but say I want to experiment with algorithm trading, what kind of APIs are available for trading? I assume they are pretty expensive for access, correct?

No need to pay for the API, just the trades you make with it -- which could get expensive.

Interactive brokers has an API for their software. It's fairly easy to use.

http://interactivebrokers.com/en/p.php?f=programInterface...

They say it right on the web page and I'll repeat it here: Try it with paper trading first! Bugs in these kinds of algorithms are costly.

+1 for IB - that's what I've been using for the past year, and they're quite solid from what I've seen. It's not a very elegant API (or trading platform [Java based + clunky] or documentation [brief at best], for that matter!), but they've got Java, C++, and Excel APIs that work, a solid selection of products, reliable performance (data kept flowing to me just fine throughout the whole mess today), and fair commissions (unless you're doing massive trades, it's usually around 2 bucks, so you can profitably offer liquidity against some pretty tight spreads). As far as algorithmic trading that's accessible to the average Joe, IB is as good as I've found. I think you need 10k to open an account, and have to maintain at least 25k to day-trade equities (a legal requirement, not an IB thing - you'll probably need at least this much if you're algo trading).

Just don't expect hand holding. You're supposed to be an experienced trader to even sign up with them (for a few products, like options, you have to pass a ridiculously easy test to trade them), and they won't save you from yourself - they'll happily let you buy 500k Euros with only $20k in your account, so be warned: if you'd made that mistake today, you'd be flat broke right now, though IIRC they do close out positions for you so at least you shouldn't end up too far in the red once your account is dunzo...

...which, to reiterate fnid2's point, it will be, unless you make sure to paper trade your damn code first! I'd also suggest that until you're really sure it's working, you never ever EVER leave it running unless you're watching it like a hawk. And code yourself a big bright red "PANIC" button that closes all your positions and shuts everything down, because you're going to end up in a situation some day where things go wonky and you just want to get out, but you'll have a dozen open positions and you don't want to have to click around to close them all.

Also, it's always a good idea to put in "everything's gone to hell" stops and place orders that time-out, even if your strategy doesn't require them (algo strategies don't typically use actual stop orders for exits): you never know when your connection might die, and you really don't want to end up stuck with nasty positions that you don't even have any way to track or close out b/c your cable went out...the positions that you hold when doing algo stuff can on occasion end up frighteningly large, which is only acceptable if you're holding them for very short period of time.

Is there anywhere we can see the amount of stop-loss orders that were executed over the course of the day (and maybe who filled them)?

AAPL, RIMM, GOOG, etc have the same pattern because they're part of the composition of the major indices, so their typical behavior is to track the major indices. If something causes the index to spike or fall, large volumes of whatever the index is composed of will traded.

Very interesting, so does that mean all those HFT programs managed to make/lose money or pretty much break even?

I interviewed at a HFT once - and one of the things that stuck out to me was that one of the guys there said "its doesn't matter which direction the market moves, just that it moves."

HFT computers identify "direct arbitrage," and make a profit off of someone else's loss.

Actually a channeling market is better for arbitrage strategies, where you are basically earning the spread.

The only real practical application I've seen of channeled arbitrage is currency markets.

They make a ton of money on rebates:

http://www.nasdaqtrader.com/TraderNews.aspx?id=ETA2009-47

Reply on news.ycombinator.com