Man versus stock picking machine
Posted on March 20, 2007 by Richard Beddard
Filed Under Ramblings |
There are still some jobs computers can’t do says Mark Thoma of Economist’s View. The Mechanical Turk was a chess machine that toured around Europe and America trouncing all-comers, including Napoleon and Benjamin Franklin. But the Turk was a hoax. The technology didn’t exist in 1854, the year of the its destruction, let alone in 1770, the year of its invention. Hiding inside was a chess grandmaster.
Fast forward 150 years and a new, young Turk. Amazon’s Mechanical Turk, billed as ‘artificial artificial intelligence‘, is a kind of anti-Google where people do ‘Human Intelligence Tasks’ - things machines can’t do. It’s like an employment agency for very, very small jobs. It’s people inside a computer, or at least behind a computer interface.
Anybody can put tasks on the 21st century Turk, like identifying items in images. Anybody can complete them and earn a fee. It made me think, but not about supplementing my income or outsourcing mundane editorial duties - an idea that obviously appealed to Matthew Stadler when he submitted this HIT:
Make my editor happy — write a better draft
Requester: Matthew P Stadler
HIT Expiration Date: Apr 1, 2007 (1 week 6 days)
Time Allotted: 10 hours
Reward: $20.00
HITs Available: 1
Description: Rewrite my draft and satisfy my editor
Keywords:
Qualifications Required: None
My thoughts turned to stock picking. Computerised stock screens are magic bullets that shoot though the information overload. Plug in a few variables and the computer ejaculates a list of recommendations.
I’m not so sure about those recommendations. Last Friday Geoff from Gannon investing interviewed me. He asked how long it takes me to evaluate a share. My reply went along these lines:
It doesn’t take me long at all. Maybe an hour or two. But finding the stocks to evaluate takes much longer. I follow 4,000-odd UK and US shares so you can imagine - I don’t have much time to spend on each. I don’t use stock screens to whittle down the number as I think they are too literal. If a company has one year of low growth in five, it’s not a growth share. That’s mad. I think the ‘fuzzy logic’ of the human brain is a better filter and the data is there on websites and in publications to page through very quickly. I spend three or four hours a week scanning data, mostly commuting to and from work.
Obviously computers and humans produce the data, but I’d rather not rely on a program to screen it for me. I’m happy for the computer to do the grunt work, but the really brainy stuff is for the Turk.
Am I smarter than silicon, or just kidding myself? After eight years of successful stock picking (in aggregate, I should say), I’m beginning to feel quietly confident. But I won’t really know for a long, long time.
I’ll link to the rest of the interview when Geoff publishes it.
Comments
9 Responses to “Man versus stock picking machine”
Leave a Reply
Well, there are proofs of such things. It has been proven that analog recurrent neural networks (humans are at least these) can compute more than Turing machines(digital serial computers). This proof can be seen in Hava Siegelmann’s Neural Networks and Analog Computation: Beyond the Turing Limit. In terms of stock/financial prediction digital computers are severely lacking. The problem is that the computers must be aware, not only of news as it occurs, but also relevant news as it occurs. I have been running into such problems in developing financial prediction software. Computers are good at giving technical predictions though, which can still be useful. When it comes down to it, I agree, it just makes more sense to have a Turk inside the machine (Plus they have such high interest rates).
The Mechanical Turk turned up disappointing results. I had to submit a “final draft” that reported the conduct of my experiment, rather than simply using what I received as a submission. That draft is online as Don’t Take Any Jobs.
The Turk has been a more satisfactory author for my personal weblog, Matthew Stadler’s Personal Weblog.
That’s brilliant. Let me get this straight. You’ve got strangers writing your blog via Amazon’s Mechanical Turk. How do I know you wrote this comment? And how do I know the Turk didn’t write all the glowing comments on your blog?
Also Kyle, thanks for your post. It’s good to know there’s a technical/scientific/intellectual basis to my ramblings
I don’t know if others have used the Turk to generate comments on my blog. I haven’t. Everything I sign my name to I have actually written. I post on “Matthew Stadler’s personal weblog” but don’t sign the pieces. I admit it is a deceitful title to give the site, but I hope it makes us question the “personal” nature of any writer’s weblog. I’m also very excited about the texts the Turk is generating. They’re much, much more interesting than my actual life, which is dull and repetitive, as most writers’ lives are.
I enjoyed your post very much. I’d have to say the same about rejecting a stock for one of 5 years that didn’t meet criteria. The machines can’t figure out.
Thanks Deborah - it’s always nice to meet a like-minded person
Incidentally I promised to post the link to the full interview on Gannon Investing. Here it is.
[...] been following Deborah’s blog ever since she left a comment on ours. I criticised the use of stock screens to pick stocks, and she’d agreed. What brought us [...]
[...] that’s tricky because you usually get what you pay for. Ranking the market can help.Although I detest screening, because it’s a way to accidentally screen out perfectly good stocks, I’m reverting to [...]