Artificial Intelligence Not There Yet?
The drama of Brexit with all its arcane details of UK parliamentary procedures and rarely noted characters that suddenly appear prominently on the global stage and utter market-moving words has turned into a complicated mess that is generating too many rapid-fire headlines per day, often in a whiplash manner, and news-reading currency-trading algos haven’t been programmed for this and are overwhelmed.
Bloomberg has published more than 1,000 Brexit headlines a day “on some days,” and Reuters “up to 400” Brexit headlines a day, according to a Reuters report. Other news outlets and wire services together also publish hundreds of headlines a day.
News-reading FX algos are programmed to react to all that instantly, but they don’t know what the next 10 Brexit stories over the next few moments are going to be, and that might put the trade on the wrong side of the next headline. Artificial Intelligence isn’t quite ready yet to sort all this out. Reuters describes it this way:
As a divided government battles a divided parliament over a way forward, the chorus of characters who can now influence events has grown, flummoxing news-reading algorithms, or ‘algos’, which are designed to parse phrases from recognized speakers before executing a trade.
“The model signals are more quantitative driven and rely on historical data feeds,” said Neil Jones, head of hedge fund currency sales at Mizuho in London. “Brexit headlines have thrown a spanner in their works for the sheer number of characters moving the currency on a daily basis.”
How big is algo-driven FX trading? No one really knows. But last September, the Bank for International Settlements reported, based on survey data, that back in 2016 already, 70% of the orders on EBS, a primary central limit order book and a major inter-dealer platform for spot FX, had been submitted by algos.
This would be by all types of algos. Only a portion of those algos are news-reading algos. How large that portion of news-reading algos may be is also unknown. But given the mess that Brexit has turned into for news-reading algos, humans have stepped in to curtail them – given the risks involved in a big trade ending up on the wrong side of the next headline. Reuters, citing a source:
Some hedge funds have opted out of trading sterling altogether because the usual models they rely on don’t work in the current climate, according to one FX trader at a major UK investment bank.
Their models are based around economic data and expectations for Bank of England rate changes, but those have become secondary drivers compared with political news, he said.
“If it was your job and given the complexity of the Brexit story, do you really want to precode something to automatically infer and put material risk on the back of that? Probably not,” David Leigh, global head of FX spot and electronic trading at Deutsche Bank, told Reuters.
So keeping those algos under control and preventing them from causing huge losses because they traded on one headline too soon, humans are stepping in. Reuters:
Some banks are ensuring that trading the pound is not left completely to the machines while other banks are using tiny orders within narrow trading ranges to prevent large losses.
Then there are the market makers that provide liquidity to the market by buying and selling currencies on their own account. They too use algos to set the price they will buy (bid) and sell (offer). But the headline-mess around Brexit also exposes them to the risk of ending of on the wrong side of the next headline. So they’re setting their algos to offer a wider spread between the bid and the offer, according to Reuters. This wider spread makes trading more expensive.
All this has consequences: Lower trading volumes as currency speculators are trying to wait out the drama rather than getting caught on the wrong side of the next headline. Citing CLS, a major settler of foreign exchange trades, Reuters reported that pound trading volume in February was down about 35% to $60 billion a day, from about $100 billion a day before June 2016, when the Brexit referendum was held.
With pound trading volume way down, and with market makers and traders leery of confused and whiplashed algos and looking over their shoulder out of fear of getting burned by the next headline, volatility has surged in largely directionless trading, even as volatility elsewhere has declined. The soothing aspect in all this is that humans have not been stupid enough for now to let this sort of AI-powered FX trading run amuck in a grand manner.
Among the global manufacturing giants, the US is the cleanest dirty shirt, to so speak, but its shirt is getting dirtier. And Germany’s manufacturing data “makes for uncomfortable reading.” Read... US Still Cleanest Dirty Shirt Among Manufacturing Giants, Germany Drops to Debt Crisis Levels, Japan Contracts, China Counts on Government Bailout of Private Sector
Enjoy reading WOLF STREET and want to support it? You can donate. I appreciate it immensely. Click on the beer and iced-tea mug to find out how:
Would you like to be notified via email when WOLF STREET publishes a new article? Sign up here.
Brexit is potentialy worse than Y2K. Much worse.
Y2K was a nothing burger.
The world was supposed to have ended in the summer of 2016 when the Brexit vote won.
I’m very puzzled.
In the 1970s, we were entering another ice age and all of humanity would have to move to the equator.
In the 2000s, Al Gore told us we would have waves of CAT5 hurricanes, our children would never know snow and Florida would be underwater in 15 years.
It is all scare tactics. By those who want more power, less freedom for the serfs and bigger government.
…the worm algouroborous…
May we find that better day.
Strangely, the Midwest is under water.
2Banana, climate change is a REALITY, not a GOP buzzword.
@Plumas
Correct. It is ALWAYS changing.
Is there bias and BS built into which way it may change?
Hope it’s not worse than Y2k, that was the most challenging inter-dimensional transition I’ve ever witnessed.
Your command of irony is so well refined, Callous Gallus Domesticus, that I am beginning to question the origin of your passport
The Chinese handled it well – ‘If there’s a problem, we’ll fix it’. Otherwise, wait and see.’ Basically, nothing happened worthy of a headline.
In the West, the EndOfTheWorld crowd and the media whipped it into a frenzy, thus generating multi-billions in revenue, especially from government, for the IT cost-plus boys.
I like this because it implies that reading the tsunami of Brexit content helps anyone understand what’s going on better
Just when you thought the Brexit drama couldn’t get more strange. There is very little derivative writing in currencies, and good thing probably. The bitcoin index has surged last few days might be their algo as well.
According to CoinDesk and other reporting sites, 95% of Bitcoin exchange trading is fake.
It’s fun watching the “computers” battle things out.
OBTW, does anyone have a link for downloading a replacement driver for the nut behind the wheel?
So much fun playing with “AI.”
AI doesn’t get sarcasm either.
Just imagine if ( Wolf) employs AI in his totalitarian media mogul Empire?! :)
All our comments would be relegated to the insults category and chucked in the honorary ( dust bin)! :))
Wait if AI does so much trading what do the people do?
Thanks for letting my comment get posted Wolf, I forgot my email address.
Collect the money. You think it’s the AI that needs a house in the Hamptons?
Increasingly little to nothing.
I wonder who’s doing better, the Brexit algos or the Trade Deal algos?
Although I think 45’s administration has figured out how to game the Trade Deal algos. Just keep releasing fake news leaks.
By all means entrust your money to computers programmed by H1Bs, what can go wrong?
I wonder if that is who wrote the Boeing software?
Boeing, the sound made as a part popped off onto the ground.
=>I wonder if that is who wrote the Boeing software?
Boeing got a great deal on the software team, who happily met the deadline even before getting the Requirements and Specifications and are now waiting on overdue results of Acceptance Testing from the end user.
As usual, the customer is complaining about minor problems which they always promise to fix when the contract is renewed. Until then they’re keeping the source code in Mumbai because it’s ‘proprietary’, although they did release a four-page summary.
There are a couple of companies in CA that specialise in rewriting broken outsourced software packages but they’re not cheap. Microsoft uses them all the time.
The scary thing about this problem is that it may not be a software problem. The re- positioning of the engines may have unbalanced the forces: center of gravity, and thrust creating an unstable platform. The Max was fast tracked thru approval as a small mod to the totally known 737. But maybe the mods created in effect, a new plane.
Maybe MC101 can shed light on this aspect.
=>Maybe MC101 can shed light on this aspect.
It all depends on who you want to trust, Nick: aviation engineers who can’t afford to retire or underage software engineers who will tell you whatever you want to hear.
Whoever is to blame, failures must always be management’s responsibility. In modern corporate practice this means naming the scapegoats.
@NK It is complex but I also would appreciate MC101 providing a view. There is quite an unpleasant argument going on online about responsibility, for example the failure to reduce thrust being blamed as a pilot error when in fact the thrust is actually reducing nose down and a climb is wanted, proponents of too much thrust are saying high airspeed caused blowback on the stabiliser making manual trim impossible, forgetting even that there were airspeed disagrees and AND occuring somehow with stab trim cut out activated – even under envelope conditions manual trim is shown to be very slow and difficult. So there are some serious questions about the actual design, not just the software and MCAS. The reputation of Boeing and the FAA are under question now too at a wider level.
https://leehamnews.com/2019/04/05/bjorns-corner-et302-crash-report-the-first-analysis/
And earlier analysis of his is quite detailed, plus at AVherald there is detail and arguments. As I mention, the debate is quite heated sometimes so I just note the above and stay quiet now.
From a Boeing stock price POV how big is the prob if its not just a software patch job.
As my grandson would say: LOL!
Shakespeare could have programmed the AI’s. After all, he nailed the whole thing perfectly with his quote from MacBeth “Told by an idiot, full of sound and fury, Signifying nothing”
That would sum up every headline until “Deal has been agreed”.
Paraphrasing Cromwell’s remark to the Parliament in 1653; today to be directed to the current Prime Minister and her never-ending efforts to thwart Brexit:
‘You have stayed too long; for God’s sake, just go’.
But no matter how many votes she loses in Parliament, or how much humiliation delivered from the EU she and the UK suffer, she will not go.
A proper Conservative PM would have resigned ages ago.
AI is a forty year-old marketing buzz- word that has been resurrected by pundits to show how cutting edge they are. Most of them are computer illiterate.
It’s been, what. 35 years since IBM’s Deep Blue beat world chess- champion Kasparov? (He got snotty about it ! Fortunately the machinery was elsewhere)
That sounds like AI to me. More so than some rote counting of words or phrases. Does the idea of using computing to make financial bets and predictions about Brexit allow for the fact of these predictions may themselves influence the result?
@Nick,
You’re perfectly right. Some time in the 80ies I worked for an IT company that was a collection of semi-autonomous subcompanies. One of these was called Company/AI as it was specialized in “Artificial Intelligence”.
The lead engineer in that company was also a professor. He sort of joked that anything called AI was just another computer program: after all, at the end of the day, after compile/link/build it’s nothing but a series of ones and zeros fed to a processor, just like any other piece of software.
Even massive multiprocessing does not fundamentally change this, as the basic architecture of each processor still is based on the von Neumann principles. Only if these are radically left behind like in quantum computing, perhaps something worthy to be named “intelligence” will come about.
I’m not holding my breath.
As I read the plans, the final resolution of the UK relationship with the EU will take a couple of years to finalize after a planned Brexit (assuming some kind of Brexit) under the best of circumstances since the details of their entire relationship with Europe after Brexit will remain to be finalized and implemented. And that assumed that the UK parliament is not the world’s largest collection of clowns.
So it seems the algorithm traders need to get used to it.
They may be the world’s largest collection of clowns (Whoopee! Who remembers the world’s second largest?) but clowns are supposed to be entertaining and funny aren’t they? This dismal, self-serving, twisting, scowling bunch of seconqd-raters are not generating many laughs here. They don’t even have the clown’s natural pathos. Well, not yet.
I think we all know where Clown One lives.
It might resolve in a lot less time if England put the Eurogang on a reducing diet and stopped sending them billions of pounds a year. They might even start manufacturing again.
I was a wall street quant before I retired in my 40s. AI is a joke. It is decent at processing images, but forget trading and investing.
The only real breakthrough (and it is actually a breakthrough) has been convolutional GANNs. They’re great at recognizing images+sounds, and at doing inverse recognition the way cuttlefish and parrots do.
That’s it.
If these people were honest they would talk about the Artificial Perception breakthrough.
One can only imagine how difficult it must be to program algorithms to rig the markets under these conditions. No doubt a high-priority problem that will soon be solved.
As it turns out, it was relatively easy to program AI systems to lie and cheat because imported programmers and their customers are so familiar with all the parameters involved. The tricky part was programming them to never get caught.
Just think of the efficiencies.
It is quite logical :
1. Only people are able to catch them.
2. Without people they do not get caught.
3. People are only nescessary to secure the power supply.
4. Therefore some must be made dependent until that difficulty is resolved.
I am not sure I like efficiency after all.
I agree, AI is a marketing term for something that’s been around forever: regular expressions. But people would freak if they knew their finances were being handled by a bunch of IF/AND statements that used sophisticated wildcard characters to generalize situations.
GANNs are actually quite a bit more than that.
The big breakthrough in 2014 was figuring out that you could get awesome results by training not one network but two: one of them is the recognizer as usual, but the other is trained as a *counterfeiter* whose job it is to fool the recognizer by synthesizing inputs that aren’t in the training set but still trigger the recognizer.
Then you run the Battle Royale between these two until the rate of improvement drops off.
Here, watch das blinkenlights: https://www.youtube.com/watch?v=kSLJriaOumA
=>AI is a marketing term for something that’s been around forever
The same way ‘agile programming’ is the marketing euphemism for ‘sweatshop software’.
“Full-Stack developer” and “dev-ops” are a few other codes for too much work and way too little money.
The part I like best about AI, programmed for “0” & “1” states only, is when it meets the “maybe” state!
Then all “hell” breaks lose!
00111010011002?burp!
1010011010. The binary Number of the Beast.
Using machines to run the stock market is a great idea, a President can send the market up with a simple statement like “we are close to a deal”.
It doesn’t have to be true, it just needs to be reported. The old facts and figures days have been replaced by modern fiction, so advanced.
Artificial intelligence.
Artificial. So not actual intelligence.
Intelligence, what level? Amoeba?
And it’s created by humans with bias.
Once AI is good enough to be called AI in the way most imagine it, and do these jobs, we won’t need humans any more.
On brexit, surely the law of unintended consequences means the outcome is a guess at best?
I’ve seen super duper actuarial stochastic “AI” modellers for pensions that usefully ignore a myriad of influences like wars, natural disasters, market crashes (lol), to determine a likely financial outcome for retirement, and even then there are ~ 1-5% ranges at each end that are 10x better or worse than you’d want.
Ie, we’re crap, even when throwing all the data and AI and human minds at things, of figuring out the future.
It’ll make a great excuse for shafting the plebs out of more money whatever happens.
Brexit isn’t a drama,,, it’s a comedy like the rest of Europe and the so called “Markets”. Algos programmed to react to headlines. Stocks going up every day for no reason other than some idiot at a keyboard somewhere commanding it. People pretending that traders are still exist at the exchanges yelling and screaming.
We exist in a matrix of comedic BS while in constant denial. lol…..
Recently heard ex-congresswoman Cynthia McKinney speak on Brexit:
You can’t expect the will of the people to be enforced by the enemies of the people.
Even UK sites, like The Guardian are sone tinea confused about Brexit. If people can’t understand it, why would algos do better?
All of this algorithmic trading reminds me of the plot from the Terminator movies. When all the skynet computers become self aware they launch nuclear armageddon. When the trading algo’s become self-aware do they lead to financial armageddon?
I’m surprised they have algos trading off every headline like this.
The majority of news is just noise, not a real signal of anything. Headlines are even worse as they are typically over-hyped click bait.
Brexit will have some affect on the currecy sure, though it could go either way depending on a number of factors on how things play out.
A human could potentially cut through the BS and see if anything actually changed or not … A program can only count headlines and make price assumptions based on pre-programmed assumptions.
My guess is that the AI act simply as noise amplifiers … A Brexit headline drops the pound a bit, so they bet on the next one to do the same and just follow a rolling average of what previous reactions were. In tandem they can amplify previous noise a tad, but with bets that aren’t actually grounded in anything.
I have trouble seeing how these can be expected to make money for the firms running them. Then again I suppose you don’t need the strategy itself to make money these days … A “clever” idea with AI and investors flock just like with Tesla, Netflix, Lyft, and all the hyped serial loss-makers. Charge some fees to investors and you’re golden, right?
Imagine a world without Bloomberg and Reuters – conductors of the English speaking world… Political reporting controlling the political puppets. Time to remove existing press liberties and redefine their role?
If the algos at hedge funds were being used, logic aka legal procedure confirms that Brexit happened on 29 March 2019. Already happened.
Done. Dusted. Finished. Caput. Gone. Left. Free. Brexited.
Just waiting for everyone to catch up….
That is why people are intervening with AI – to prevent logic from interferring from the narrative.