Renting a Supercomputer

Dont like the basketball analogy, fine...

Your playing chess against Kasparov.
It took a HUGE team of engineer, super computing
power, and decades of reasearch to finally beat this guy.

Are you going to learn chess, or do you plan to
put that kind of effort into developing a chess computer
that can beat the best chess players?

Do you think actually being a good chess player
could help you in any way?


I can tell you this.... it would take me 1/20th the time
to write a profitable full auto system now, compared to when I
knew very little about trading.

peace

axeman




Quote from aphexcoil:




This isn't basketball -- this is a game where the brightest minds are playing. I would like to do my part in trying out a system that covers all bases -- even if it took me 5 years of designing such a system.
More...
 
Quote from aphexcoil:
Whoever sold me what I bought did so because they believe the opposite of what I believe -- why else would they sell?
More...

People don't always sell because they feel the price is going down. It could be a hedge, an arb, an exit via profit-target, one leg of a synthetic, etc....

Also, I don't know why you "need" a supercomputer. You mentioned "a few billion ticks" worth of data. Have you even defined the problem yet? The last 5 years of ticks on SP is only 5M ticks. On ES it is 8M ticks. 30-yr Bonds 1.5M ticks. (This is using active contract).

Also, there is no point computing at the tick level if you cannot compete at that level. In other words, you may find a pocket of inefficiency to give you an advantage... but say that pocket only lasts for 5 ticks. If you don't have the infrastructure and automated system setup to take advantage instantly then why even bother? Move up to 1m or 2m time frames where you can compete. btw, arb houses do this very thing on tick data with automated systems.

Regarding compute power, I number crunch on various NN models with GA's every day. You don't need a supercomputer. A Pentium 4 has much more compute power than all of the computers the Predictors were using in the book.
 
Quote from dottom:

Also, there is no point computing at the tick level if you cannot compete at that level. In other words, you may find a pocket of inefficiency to give you an advantage... but say that pocket only lasts for 5 ticks. If you don't have the infrastructure and automated system setup to take advantage instantly then why even bother? Move up to 1m or 2m time frames where you can compete. btw, arb houses do this very thing on tick data with automated systems.
More...

So we again come down to:

Who's got the fastest connections with lowest transaction costs.

Sounds like a reoccuring theme.
 
Quote from dottom:

Also, there is no point computing at the tick level if you cannot compete at that level. In other words, you may find a pocket of inefficiency to give you an advantage... but say that pocket only lasts for 5 ticks. If you don't have the infrastructure and automated system setup to take advantage instantly then why even bother? Move up to 1m or 2m time frames where you can compete. btw, arb houses do this very thing on tick data with automated systems.
More...

I agree. Start with 2, 5, 10 min. time frames and see what you can get. There's so much noise in tickdata, I guess noise=trends at such a short time frame... plus slippage, commissions - you probably have no chance to win at this level.
PS. I use old Celeron 333MHz for backtesting with 5min bars (1 year of data) and it works fine!
 
Quote from aphexcoil:


So we again come down to:

Who's got the fastest connections with lowest transaction costs.

Sounds like a reoccuring theme.
More...
Only if you're trying to take advantage of arb-type inefficiencies lasting for a few ticks. You can throttle your analysis time frame back to 1m or 2m or what have you and still find pockets of inefficiencies that a retail trader can exploit.
 
Aphie... Aphie... Aphie...

It's not a flame so take this seriously. It's great that you're looking into these matters with computerized trading. I understand with what you are saying.

I don't know how you are trading right now. But if you're still having a hard time trading and looking for a way out this with all the neural net stuff, you're looking at the wrong direction.

What you are doing is just going around in circles and it's not going to take you anywhere as a trader. If you want to overcome you psychological problems, first you have to fight it, just like you have to be in the market to profit from it.

Just my 2 cents. I just hope that you're not going into the supercomputer stuff for the reason I mentioned.

Other than that, great thread. Very informative.
 
posted by aphie

"That's why I am trying to understand the application of neural-nets. The most successful application of a program would be one that does not have set code to look for patterns but one that can "learn and rewrite itself" as it observes. The only thing I know that can do this is neural nets."

And so we're back to round one...bud, "one that can learn and rewrite itself as it observes"...IS YOUR BRAIN.
This is what discretionary trading is all about, the human brain which is itself a neural network. You have all the hardwired tools inside the cranium on your neck so USE IT TO OBSERVE THE MARKETS, TRADE, AND LEARN/REWRITE AS YOU OBSERVE.

and if you respond with a "but my brain can't process a billion ticks of data" i'll kick you in the arse...of course it can't, and it doesn't NEED to be able to in order to make money in the market.

you are currently flying on a wing and a prayer, hopefully you'll wake up.
 
Quote from aphexcoil:



So we again come down to:

Who's got the fastest connections with lowest transaction costs.

Sounds like a reoccuring theme.
More...
Yes,

but unless you are metooxx, I STRONGLY don't recommend this route - the resources required are IMMENSE.

nitro
 
you are thinkin way to much - i just go in every morning with an open mind , read the market environment - buy the shit i think is gonna go up and short the shit i think is gonna go down -then cash out after i make $5k to $10k . see, easy!
 
You must log in or register to reply here.
Back
Top