Neural Nets and Genetic Algorithms?

It's like a bunch of Rednecks talking about politics in a bar or...

It was in the "Jokes" about a guy trying talk about Nuclear Power or something like that in the airplane...

I think this... and I think that...

But...

All AI is stupid... hey God made us and we're stupid from his perspective (probably).

People made AI, they're the subject of a human and logically cannot and does not hold the vital component to surpass our intelligence.
 
in reality,

AI and NN (artificial intelligence and neural networks) both describe aspects of the human mind, however the human mind has intelligence and its not artificial, and the connections that we make when we learn a concept or reinforce a concept through confirmation of, say a chart pattern are the brain (neural) networks (series of connections of thoughts) that occur when we think.

the best NN is our own brain because it also has the compute power to load up two other essential programs that must run in tandem with the Analytical Program, and those are:
1) emotional state - that allows us to actually commit to a trade
2) risk manangement - that takes into account our physical needs for profit, and psychological needs for being confirmed as correct in our choices of trades or timing

no programs have successfully automated those other two portions to complete the cycle in automated trades entry, only to some minor extent, and certainly not to be trusted with futures trading, only the slower equity trading counterpart.

NN"s are brilliant in concept, theory, practice of programming and leave far too much to be desired in real practice.

train your own neural net and it will work far better than these automated kinds
 
Quote from limitdown:


train your own neural net and it will work far better than these automated kinds
More...

Indeed. Even if you use a programatic NN, you would still want to use your own head to process whether or not you act on the information provided.

The brain may be able to be better at figuring out when to trade or not trade (computer programs couldn't accurately predict 9/11 or similar catastrophic events).
Also the brain is obviously good at analyzing a single stock and determining when to buy and a particular strategy.

Where the computer programs come in is the ability to do it quickly and at scale. A distributed network of computers could easily crunch through and analyze all the stocks to determine a good trading strategy - whereas for a single human to do that would take too long to be practical (by the time the task was completed, the resultant decision would likely no longer be the best fit).
This is my main interest in brining computers into play - allowing them to do the dirty work on a grand scale and point out a narrow range of possibilities. Then from there, I can decide myself what to do with that information.
 
As long as there are PEOPLE TRADING you cannot and will not predict movements before the fact 100%.

I'm surprised that "Chaos Theory" is not on anyone's reading list. It has a couple of things that are relevant to trading:

1. Trends exist, BUT they can reverse themselves quickly with no prior warning.

2. In order to predict with a very high degree of probability you have to know as many variables as possible that has an influence on the item you're measuring. Will the butterfly wing's flapping in China effect in some way the hurricane brewing off the coast of the US?

I originally started my career in forensic psychology because the human mind is complicated enough that it surprises the hell out of me that the majority of us exist "normally". So, I originally thought it was more of a challenge to work with those whose mind was wired differently and could not exist outside of a prison. Oh, and figure that out and save the world from criminals while I was at it.

Black box systems, in theory, should have taken the place of more personable traders. The mind is complicated and works in ways we cannot even imagine and works the most when we're not even thinking about it.

If black box systems work accurately, then why does Bright Trading have traders in offices and why have they not programmed the strategy they teach to a computer? And why are there unprofitable traders who have been taught a system?

I think that no computer can take into effect certain variables, like emotion, measure it and trade with a high degree of accuracy.
 
Simple answer: because the time has not come yet but with low cost computers and internet sure the trend can only accelerare. In fact a large percentage of trading volume on the Nyse comes solely from automated trading. It only concerns firms but probably more and more traders will join automatic trading. Even if you are not for competition will force you.

Antropomorphic vision is always blinding so remember that a dumb machine has beaten Kasparov at Chess:

http://www.gamesdomain.com/gdreview/depart/may97/craig.html
Exerpt:

"The chess-playing community is also shell-shocked. Some are convinced that Kasparov took a payoff to lose the game on purpose to generate more interest in Deep Blue. Others say that Kasparov's style played right into Deep Blue's hands and that Karpov would not fall into that trap. Or that if he'd simply tried harder he'd have won. Yet the fact is there that he lost, and lost for the first time in his life. Kasparov could not psych out his emotionless opponent; opposite him sat one of the IBM team. Deep Blue hummed away in another room, distanced from the action. Nor could Kasparov attack the weaknesses in his opponent's style of play, for Deep Blue has no style, and if anything represents a hybrid of many observed styles. Kasparov is rumoured to have practised against a number of computer opponents to try to get a feel for the style he'd face, but if anything that probably did more harm than good. Inevitably the odds were stacked against the unfortunate human sacrifice, and Deep Blue accepted the offering in cold blood.

The bottom line is that the number crunching ability of the 32-node IBM "supercomputer" has finally outweighed the natural ability of the best human player."

Quote from facultus:



If black box systems work accurately, then why does Bright Trading have traders in offices and why have they not programmed the strategy they teach to a computer? And why are there unprofitable traders who have been taught a system?

I think that no computer can take into effect certain variables, like emotion, measure it and trade with a high degree of accuracy.
More...
 
I agree that computers have a place for grinding through all the data to find "inconsistencies", but there will come a time when an event will happen that no model exists for. Granted, there may be enough time for profits to be taken to accept a loss for the time when the event happens.

Anyone remember Long Term Capital Management?

 
Well, unlike markets chess is not a stochastic system. In a simpler game, checkers, computers have been world champios for over 10 years. On the other hand, I would not envision a good speech recongnition system or a machine translation system for the next 50 years or so.

NN-based black box systems for stat arb have been making money for a while, but there is no prediction involved, just the issue of timing. I have overheard among equity folks that there is someone who is doing basket trading using NN, but I would not stake my $$ on that fund.
 
Of course there are some NN that makes prediction, in fact it the first use NN are normally used for. Non linear stochastic models like Garch-like models for example. If you want to use NN for rules based system it is possible but there are other techniques of IA for that. And that's the common point with Chess: some trading system is like Chess based on rules. If I take my case my model is a non linear model which could be optimised by an NN, but the trading system would use rules and so rather classifiers or expert system than NN per se. One must not confuse model which are mostly complex and rules system which can be from very simplist (not even based on a model or at least so primitive that it wouldn't be called a model) to very complex ones. And unlike chess trading can suffice with very few rules if not one must be superman to be a trader. If some find that becoming a good trader is hard, I don't think that it is as hard as becoming a good chess player, I would rather say the contrary.

Quote from sle:

NN-based black box systems for stat arb have been making money for a while, but there is no prediction involved.
More...
 
The problem with traders are like with workers when the Boss announces them that they will automatise everything. Of course there still need humans but for surveillance only. This is the trend of productivity, and Wall Street that is used to acclaim better productivity and sacking of millions of workers will probably do the same when the bankers will say that thanks to automated trading systems they will economise some (high) costfull traders, don't you think so objectively ?

When this time will come some traders will feel in the future what a worker can feel when he loses his job. Then you will think twice before acclaiming to such announcement of "productivity" gain. Workers that have adapted have found a new job others had stayed jobless. So should be for some traders.
 
You must log in or register to reply here.
Back
Top