Hedge fund robots crushed human rivals in 2014
cnbc.com
cnbc.com
This is headed for 2008 all over again because we've learned nothing. These algorithms rack up consistent wins for years on end, the markets rise and rise, everyone gets bonuses and all is great. Then the market turns on a dime, all the computer algorithms break at once because chaos theory takes over and your trend algorithms are useless. Liquidity disappears in an instant and the the federal reserve steps in and gives away free money to the banks until the banks become solvent again. This will continue until the parasite kills the host.
Edit: No, child posts, you cannot make the leap that trend following algos are the same thing as modeling MBS risk (which isn't an algo...).
One of the major failing of modern statistics instruction is 6+ sigma events don't look like 1-5 sigma events. Case in point global death totals per second probably fit a very well defined curve, but Hiroshima and Nagasaki where well outside that curve.
The flash crash was the result of bad algorithms, the tools acting on their own without significant human guidance or erroneous input. But that was a relatively minor event that temporarily affected professional market makers. Something like that could happen again, with a much larger impact. But the 2008 crash was not it, and wasn't related or even similar.
It is difficult to predict 6+ sigma events, but humans as a whole are no better at it than algorithms are. There were plenty of people who saw things that others didn't and made big short bets against the housing market. The problem there is that everyone else didn't listen to them. Eventually we can improve statistics to catch more outliers. Good luck improving human nature to get people to listen to predictions that go against the herd.
And if someone really wants to place blame, it's pretty tough to ignore the regulatory and interest rate changes that created the environment that allowed it all to happen. But again, like you said, good luck improving human nature.
Banks have shown themselves to be pretty irresponsible with (practically) free credit though. That's a mistake we shouldn't repeat again.
Often bad nails are the direct cause of a collapse. As in someone used 3 inch nails when they should have used 6 inch nails, or they were not rust resistant etc.
In the end Root cause analysis is about looking past the direct cause to keep from repeating the same mistakes you need to look at both the direct case and dig further. In this case you look at algorithms misclassifying a lot of bad debt and then pealing back the layers. In the nails example it might be someone incompetent designed a foot bridge or someone tried to skim money by using 3 vs 6 inch nails or whatever.
In the bridge example replacing the crooked supplier is all well and good, but you also need to fix all the other bridges made with 3 inch nails.
PS: Where to place blame and the direct cause are far from the same thing.
Why wouldn't trend following be the winning strategy when the market stayed on a steady climb? Anyone who hasn't continued with that strategy, attempting to call a top to this current market, has probably taken a beating many times over.
If anything, Algos helped to correct the market more quickly as those who chose poorly were washed away with greater speed.
- a regulatory environment encouraging subprime loans
- high yielding securities which were rated as AAA (by a legal oligopoly of bond rating agencies) and thus could be used for bank reserves
- a short term interest rate that did not respond to market pressure (because determined by the fed) and thus created a gigantic carry trade
Mistakes in risk models always happen, and they can be caused by many things, including an over-reliance on mathematical model. However, for a real catastrophe, you need to remove all the feedback mechanisms such as
- a free market that would allow the short rate to rise with increased demand
- a free market that would allow bond rating agencies to compete
- a free market that would allow bank to compete for balance sheet quality rather than having a government insurance scheme creating a race to the bottom
Also, one big issue in 2008 is that no one thought so many different asset classes would be correlated together and so many were caught out when the market devolved into just 2 meta asset classes : "risk-on" and "risk-off". Can the algorithms account for this?
Can they also account for the fact that we have never seen such a persistently low interest rate environment in so many major economies at the same time? Or is that off-model? Looking at their human counterparts, a recent article speculated that it has been going on for so long, and there has been so much turnover at the financial institutions, that there may be many young traders who literally have no experience of what happens when interest rates are put to more normal levels.
Things like the major dislocation from the Swiss franc devaluation show the fallibility of trend following models that fail to capture the known-unknowns:
"When the “off-model” event was the breakdown of parts of the wholesale money market in 2007, their surprise was just about forgivable: in the case of the Swiss revaluation [which even Goldman Sachs called a '20-plus standard deviation' occurrence], to have failed to visualise the possibility is rank incompetence [2]."
-- [1] http://abcnews.go.com/Business/story?id=4842282&page=1&singl...
[2] http://www.ft.com/cms/s/0/5a06ef16-b5e4-11e4-a577-00144feab7...
Wrong. Trend followers did extremely well in 2008-2010. As the article says it was the years 2011-2013 that they struggled with.
My question would be, what happens to the markets if there is a quick fall, the algorithms shut down, and all the HFT liquidity is removed?
Not all of them have kill switches - Knight Trading had something with no kill switch that cost them $400M in a matter of hours. (It wasn't even a trading program - some kind of simulation system that was erroneously allowed on the real market).
You could say "it didn't need a kill switch, because it wasn't a trading algo!". But many trading systems are made of small parts that can break, and you only know your kill switch works after it saved you - otherwise, it's just an "untested feature before QA".
> My question would be, what happens to the markets if there is a quick fall, the algorithms shut down, and all the HFT liquidity is removed?
We've seen such mini events before. We see what's known as a "flash crash" (price goes down significantly within seconds, only to come back to more or less the same place), or alternately, a "flash smash" if it goes up (to come down later). This has been happening since 2007 at a frequency of about once every two months. Sometimes trades happening during the flash {sm,cr}ash get retroactively canceled, and sometimes they do not.
People are pissed and jumping ship. There's way more people than there are positions available because the industry is contracting/consolidating.
Sure, the algos performed well this year, but these are the firms that survived. Lots of players got crushed playing the algo game and are no longer willing to play.
Some are living off their savings for a little while and seeing where things go in the field before trying to return. A few have left to work at some startups. Most are planning on staying in finance though.
This probably isn't obvious to people outside of the trading world, but Newedge is a massive futures broker. This article is basically a puff piece quoting a big futures broker telling you how great it is to trade futures. I wouldn't read too much more into it than that.
(1) Don't know if this includes fees (2) Don't know if this includes dividends
"Financial markets used to be about getting money into productive enterprises, not out of them"
Right on. Let all the leveraged bots and algos devour each other. Sooner or later "real economy" will decouple anyway. And with the savings of millions of individuals, pension funds and insurance corporations ping-ponging between botnets, the real "hedge" will be to remain liquid and competetive outside this hodgepodge when it all inevitably ends up belly-up.
I have never heard a single good argument against HFT.
HFT does not make ridiculous amounts of money these days - and the money it does make is at the expense of large institutional investors - e.g. Goldman Sachs, JP Morgan, et al. Are you so worried that a bunch of clever young mathematicians are taking money away from those poor old put-upon Morganites? Oh how dreadful. HFT means tighter spreads, which means cheaper and more efficient trading for everyone - i.e. your individual workers investing their 401ks, and so forth.
People's arguments against HFT seem to normally come up to '...because computers!'
Do yourself a favour and read this:
https://scottlocklin.wordpress.com/2014/04/04/michael-lewis-...
There are two problems with HFT being carried on by computers. One is obvious: HFT trades done by computer happen super-fast and arbitrage away any slack in the system between bid/ask prices and so on. There's no way a human trader can compete, which means that after a human has made the mental investment to understand how such trades work, s/he is unable to exploit this knowledge and get a foot onto the financial industry ladder. This is aggravating to people who started out trading small-cap stocksor suchlike. Now, day traders are notoriously bad investors to start with and it might be that the market is better off without them, but there's a big qualitative difference between being a bad day-trader who's competing against other humans in the same market, and competing with a machine whose owners have paid extra for it to be located close to the exchange to shave a few milliseconds off execution time. It's like if you put a super-fast mechanical runner in one lane of the Olympic 100 meter dash; while objectively it should have no bearing on the performance of the human participants, which is what we aim to measure, our psychology is not so logical, so seeing a robot complete the 100 meters in, say, 8.5 seconds (handily beating any human for the foreseeable future) would have a massive demoralizing effect on both contestants and spectators. It's one thing to know in the abstract that a machine is much faster than you; it's another to see your best effort radically devalued.
A second reason is a bit more abstract and rarely articulated, but IMHO equally important: human trades are ultimately motivated by human desire. Speculators want more money, but long-term investors want yield to provide a sustainable income, commodity buyers want commodities as industrial inputs or for resale to the public, and people buy commodities at retail because they need to eat, build shelter, and so on. The market ultimately exists to serve human need, notwithstanding the fact that it often serves human greed (which is just hypertrophied need and often correlated with some psychic deficit). HFT and algorithms in general work blindly; as very basic software machines, they operate purely deterministically, with no sense of the context within which their operations take place. Of course, human traders can also lose sight of context (I've heard cocaine is great for this :-p) but even the worst excesses of leverage or psychological dynamics are constrained by the social structures within which they take place; human inefficiency (in decision-making, and in implementation of that decision by others along the chain of execution) serves as a buffer to limit excess, although this is very much a function of scale - see the historical excesses in totalitarian states, for example.
Getting back to HFT and algorithmic trading, what people worry about at the fundamental level is that as it improves over time and more and more money is under the management of computers, none of which have any moral or social context for their trading decisions, our social buffering against excess becomes less and less effective. This is also a reflection of anxiety about late-stage capitalism; logically, we sense that in a purely deterministic market we would eventually see more and more capital concentration similar to the inevitability of a single winner in a Monopoly game, and that's a stressful prospect because we know that such circumstances are not socially sustainable.
The problem with modern financial markets is that they have become quite abstract. Finance is maths, which means it's code, which means you start abstracting away from the implementation to get to a high level. Just because it's not immediately clear to a layperson what's going on, doesn't mean it's not useful or real.
Imagine what would happen for most programmers if you tried to explain to a layperson what you do.
"I'm a developer"
"Oh, you make websites?"
"Well, uh, no, you see, I make tools for making it easier to make websites"
"Oh, like what?"
"Um, well I make a compiler for turning one language into another"
"Why?"
"Well, when you make a website, you have to do all this stuff, but, well, you could do less stuff, just more clever, and then it's easier and you don't have to worry about for loops and such"
"What's a for loop?"
"Oh that doesn't matter. But now we can use a monadic map instead, which is much easier, developers don't want to write loops, map is better. But Javascript is a bit fiddly, so now I've written another language, so you can write in that, and then you run my tool, and then you get Javascript, so the browsers can understand it."
"Oh, so the browser only runs Javascript?"
"Yes. Well no. Actually most of them run some kind of JIT compiler, which turns javascript into another language, and then they run that"
"So your language becomes another language which becomes another language? Why not just write the other one?"
"Well, because that's not really what we want to do, it's awkward, and inefficient, and bug-prone..."
"But your language isn't a real thing? No-one's actually using it? My computer never gets it, never runs it, never understands it? Shouldn't you be doing real development?"
I'd be more worried if several humans were significantly outperforming algorithms over a significant amount of time because that is an unlikely outcome in an honest market.
Automation will also drive down costs.
You can flip a coin every year. Heads; double your money, tails; lose your bet. You can easily win 3 years in a row (12.5% chance), but getting 800% return on that risk is literally worthless in the long run, as the chance of winning or losing is equal.