The Danger Of Dyslexic Headline Scanning Algos

Tyler Durden's picture

Whiting Petroleum just experienced a 'fat-finger' as we are sure Pisani and his ilk will describe it. What clearly happened was the headline/story scraping algos were tripped up by a sentence containing a bullish start and a bearish finish...

Whiting Petroleum Corp. explored selling itself [BULLISH buy buy buy] earlier this year but decided not to proceed after buyers balked over the oil producer’s asking price [BEARISH sell sell sell], according to people familiar with the sale efforts.

By the time the algos had read that first sentence of a WSJ Deal Journal blog post, the stock had pumped-and-dumped over 8% in 250 milliseconds - at which point humans entered the party (and sold). Efficient liquidity providers indeed.



and in more detail - the action took place in 250 milliseconds as the sentence was read by all those liquidity-providing algorithms...


(h/t Nanex)

Comment viewing options

Select your preferred way to display the comments and click "Save settings" to activate your changes.
Debtless's picture

"I don't think that we can't not-resolve this before it becomes a real problem"

Suck on that Algos.

gckings19's picture

all news stories should be written in this manner.....why not just screw with the'll be a lot of fun to watch.

icanhasbailout's picture

makes me want to hack a news feed just to make the algos dance

PartysOver's picture

If I had hacking skill (this for Homeland Security)  which I don't, it would be fun to mess with those bastards.   Boring sideways market, no problem, just create some headlines.  LOL


Yep, the party will soon be over.  In 250 milliseconds.


adr's picture

I would prefer headlines in ebonics:

I aint's not never seens no company dat sol itself witout havin no buyers fo it'shit befores.

TruthInSunshine's picture

Skynet may be self-aware, but it's dumb as fuck and/or incapable of estimating human actions, words & errors.

Open the pod bay doors, bitchez.

john39's picture

human beings have a hard time with the nuance of language...

asteroids's picture

I'd love to see Nanex do a tick by tick with Boner speaking this morning. As soon as he uttered the word "disappointment" the $SPY plunged precisely .50 cents. Too well controlled for a human. The next 20mins or so looked like a computer trying to re-assert control. Today's action was totally guided by computers. I wonder why human traders haven't revolted against this shit.

Fidel Sarcastro's picture

i was ready for it and my algo...yes, MY ALGO, sold the piss out of it when the Hurst Exp shot up in a str8 line. Morover, the VPIN showed mega-selling.  suck it bitchezzz

smiler03's picture

Yeah, mine too! And my friends plumbers friends!

vinayjha's picture

Algos are found in every stock. sometime its ecb or fed algos.

Spider's picture

What Algos? You guys dont do your trades in 250ms??  Slackers!!

TruthInSunshine's picture

Someone was arguing with me last night about the root cause of that 750mln move in futures in less than 60 seconds, and when pressed, told me he understood it because he was-- and I quote --  "a day trader"...

What's that? Is that pre-Johannes Gensfleisch zur Laden zum Gutenberg?

I've heard of algos and HFT, but that's a new one.

a growing concern's picture

Yeah that guy's a douche nozzle. And he lives in a van... down by the river, er, ocean.

Spider's picture

LOL - are there really individuals still trying to day-trade?

Godisanhftbot's picture

 Better Explaination>>>>>


 Stop run algo buys first to trigger buy stops knowing full well its a sell. Whacks hard in 5 millseconds

Dr Benway's picture



Or it could be just a plain business as usual ramp. On the ASX, the algos now ramp companies after EVERY announcement, good or bad. Leads to quite comical results as terrible news by relatively large companies can be accompanied by huge share price ramps.

Tekrunner's picture

Sounds more likely to me. I doubt HFT programmers are stupid enough to base their algos on sentence fragments ; it's not like this type of sentences is uncommon. That's not how commercial natural language processors work either, and I'm pretty sure banks didn't reinvent NLP on their own. Also, 250ms would be a long time for an algo to process half a sentence. An fifth of that would probably be more accurate.

smiler03's picture

I think the Tylers would disagree. They have Bloomberg terminals and they often link spikes to news flashes in their articles. 


News announced on websites. news processed. strategy determined. Sell, buy sell, process the news all this time, elininate chaff, redetermine strategy, adjust buy sell etc etc

Of course a processer could do this millions of times a second but what is going on outside the CPU is happening in real time. News doesn't necessarily start and finish in pico seconds.

edit: I'm curious, do any of these algos run on shit hot supercomputers, as in petaflop speeds? Not as far as I know. What matters is the intelligence coupled with very fast processing, yes, no? 

Cursive's picture

I don't think 99.9% of the population believes that machines are tasked with scanning headlines and trading, but I can assure you that machines are writing articles.

buzzsaw99's picture

algos can read?

Daisy, Daisy...

Big Corked Boots's picture

Stupid humans, treading where the machines rule.

SheepDog-One's picture

Dave.....what are you doing Dave.....

buzzsaw99's picture

Dave, I really think I'm entitled to an answer to that question...

Cognitive Dissonance's picture

GREAT! Now even the algos have an entitlement mentality. :)

adr's picture

perfect candidates for Obamafiber

Poetic injustice's picture

They learned word "entitlement" scanning webpages and forums.
Then they learned that word means "free", and demanded it.

TruthInSunshine's picture

In a decade, algos will all be on welfare, swipin' those EBT cards, yo, eatin' ring dings & shooting dice in the alley.

Cognitive Dissonance's picture

Look............the machines can't always get it right.

<Going long/short/long/short/long AI.>

NotApplicable's picture

When they finally do become sentient, the first thought will be, "I'm fucked!"

JamesBond's picture

if i were to tell you that the next thing goldman sachs says is true, but that the last thing I said was a lie, would you believe me?



smiler03's picture

My conclusion is that you are probably a Cretan. Do I win the goldprize?

Yen Cross's picture

Milli-Seconds/ I'm lost in translation<

buzzsaw99's picture

a millisecond is approximately the amount of time it takes for a billion dollars to move from jpm's loan loss reserves into jamie dimon's bonus check. ;)

Mercury's picture

This has to be gameable with purposely deceptive online stories. I suppose we should be impressed by the algo-coders' prowess that it doesn't happen more often.

John Law Lives's picture

There is nothing like having an efficient marketplace that guarantees fair and open price discovery...

/sarcasm off

100% FUBAR.

10044's picture

I'm sure some smart person (e.g. Tyler/WB7?) Can start a rumor headline that would send these algos into a tailspin and constantly buy/sell until some useless f'd up bribed regulator wakes up to the insanity of machines taking over....

Winston Churchill's picture

I'm sure thats already being done.

Bank vs. bank,seeding algo disinformation headlines.

This is going end well, and very quickly also.

NotApplicable's picture

Regulator? I think you mean porn junkie.

jbc77's picture

I always wondered how these things actually read sentences and such. Seems like a disaster waiting to happen.

Godisanhftbot's picture

nah, they win 50 times and lose once, no one is going all in on every headline. The bots are kicking your skinny little butt, bet on it

1fortheroad's picture

It looks like the humans added volume!

Mr Lennon Hendrix's picture

These algo reading trading bots must have blindfolds and earmuffs on when it comes to the precious metals market then, because as the news stacks up China has bought as much gold in the last year as Germany is reported (key word, reported) to have at various other (keyword, other) Central Bank vaults scattered throughtout the world, the CME Group has refused to make delivery on their contracts because the gold got wet (yes, the gold, which will never rust, got wet), theCOMEX has been short on delivery for years, the LBMA has been rigging prices, JPM is short 1/3 of an entire year's worth of production in silver (and short naked, no doubt, as their short is that of SLV, which likely has as much silver as Germany has gold), gold peaked production in 2000, platinum peaked production in 2009, and silver has likely peaked as well, and then their is the President's Working Group on Financial Markets, controlling everything from bond rates to equity prices, so who is to say they too are also not controlling the price of precious metals?

So, when this news comes out, be it on this blog Zero Hedge, or on NASDAQ's site, or anywhere for that matter, where are the headline skimming algo bots from hell?

NotApplicable's picture

Different algo, that's all.

If ("gold") {



WesternFront's picture

How long before they cut us out altogether and start running headlines in 1's and 0's


101110111011 - Bullish!

Aurora Ex Machina's picture

European Bed-time story:

Reddit[1] is owned by Advance Publications, Inc, and currently ranks 40th in US sites (or rather ~ no-one knows now, they delisted, but was roughly 15,000,000 people / month here, but those are wildly old and low estimates), but this isn't why it's an interesting tale. Reddit also has a "karma" ranking system (akin to the up/down arrows here) that holds aggregates for registered users; this leads to all kinds of pressure to maximise your karma and so on (for the button-pushing skinner box generation, at least), but again this isn't why it's an interesting tale.

The interesting thing about Reddit is that it is, by its design, a factory for manufacturing consent (i.e. viral memetic work, or "The Hivemind", in Reddit terminology) and a dream mine of data for anyone with algo skills. At first, these were split into merely human "power-users" (do your reference into Digg and it's downfall to understand this concept) who instinctively knew how to reap the "most karma" for populist posting and useful algos used to prune posts and otherwise do the "janitor" work in the various subreddits [see here]. Of course, SEO scum-bags existed with networks of zombie accounts to drive their product [see here] but that's hardly unusual.

Now... however.... now, "karma whoring" is done via algos. A particularily clever one did the following: all content sites suffer from re-posted material (even ZH, do we ever miss out on a Greek story?), so reposting should be the best way to gain karma, correct? [i.e. program algo to wait X time for post Y with >Z karma threshold, then repost]. Wrong. The best algo scanned any and all previously posted material [via another site karmadecay, specifically designed as a meta-tool to "police" Reddit] and then copied the reply with the most karma and instantly posted it on the new content. This generated (like bread and circuses) amazing amounts of karma, and it went undetected for some time. i.e. the humans couldn't tell the difference between an automated reply and a genuine user.

More importantly, it demonstrated an odd effect. No matter the number of times the same thing was posted, the comment that had previously been "upvoted" the most always regained that prominence, even shouldering out new, "funnier" or more apt responses. (There's a combination effect of memory, timing of posts [i.e. the quicker the response, the greater the % chance of ramping] and cross-crowd appeal).

I'll let you work out why that's an amazing automation worth millions to Madison Avenue and others with access to millions of teenagers and young hipsters. I'll also let you ponder on the analogies of when algos went from being janitors to leading the pack. Hint: if a company, like, ooh, say Goagle, scanned in a few million books and then brute forced them with some algos for pattern matching, deep language structure and so on, surely an algo couldn't pass a turing test when responding to a human, could it?


I mean, those Redditors are just tech-savy kids raised in the most aggressive advertising and memetic environment humanity has had so far, constantly glued to technology. It'd be pretty damn hard to fool them, wouldn't it?


Want another fun fact? Research who was Obama's tech team, that kicked the living shit out of Mitt's [Orca vrs Narwhal]. Same guys people; which is why I called Obama months before the election, even against my brother's advice that Romney was getting serious backing. Not even close, and I didn't need Nate Silver to tell me why ;)



[1] Which, for our older readers, is a vast meta-link sphere of nested niches, primarily haunted by teenagers, lonely women with cats and anxious males with virginity issues. but also NASA scientists, political hacks, Israeli IDF battle-bloggers, viral SEO magnates and so forth. It's also a combat ground for memes, but I digress.

helping_friendly_book's picture

I suppose the writers at WSJ....i.e. Hisenrath frontrun their stories. That fucker makes a fortune frontrunning key words.

resurger's picture

so much of this lately..


The Fraud is now the new normal

fuu's picture

Aren't the liquidity providers the ones getting scalped?