Stock market prediction stats

[Nathan]

If you think I'm reading all this, you've got another thing coming.

 

[FatRat]

...but I have a 10% over 21 days running as well just out of curiosity.

Ohhhhhhhh, now we're talking, cheeeerching!

Like you said before, its really all about giving the slightest advantage. The great thing is that they give you a place to start looking...

 

[notoriousQQ]

my head hurts from reading that.

 

[NoDaddyNo]

Just had a chance to write up some code to check out how much could be made with this (on paper).

Assuming a $20 trading fee (on buy and on sell for every stock purchased).

Just pick a random stock out of the N stocks given to you and put all of your money that will evenly go into it and still pay the trading fee.

Then if the time period hasn't been exceeded, check to see if it has gone up over the percentage required.
If it has gone up enough, sell it all and then the next day buy a random one.
If it doesn't go up enough during the time allotted, then sell all on the last day at whatever that price is.

Repeat again once all are sold.

For the 1 percent over 5 days:
Return on $10K: 10572.82
Return on $50K: 52872.39
Return on $100K: 105747.11
Return on $250K: 264370.26
Return on $500K: 528741.87
Return on $1M: 1057486.05

For 2 percent over 21 days:
Return on $10K: 10759
Return on $50K: 53807.52
Return on $100K: 107618.01
Return on $250K: 269052.45
Return on $500K: 538107.87
Return on $1M: 1076219.35

While that does result in profits, I think it is obvious that is not the ideal strategy.

Now I'm working on the test code for a different strategy and will post that up once I can confirm that its so far apparently crazy returns are in fact legit (they come about from compounding interest).

There are obviously many different ways to trade - I'm just trying out a few obivous and relatively safe ones.

 

[jnuts]

You doing Neural on this? If so, I've got some questions. I'm doing some ANN work, but not in the economic realm.

 

[ZKaudio]

if u need any coding / design assistance i have some friends right up your alley.

 

[NoDaddyNo]

Jnuts - I have actually moved a bit away from neural net stuff (standard backprop stuff), but I am very familiar with it and have certainly tried this in that area before.
As I have said before when it comes up, the neural nets tend to make more predictions, but aren't always right.
The current system I have gives a high penalty for being wrong and would rather not give a prediciton at all that be wrong.

It is interesting because there are a ton of variables that can be tweaked and it is already tweaking *a lot* of them in an automated fashion when it learns (over 50million calculations per stock - that is actually likely a low estimate).

And ZKaudio, thanks, but I am a design person with 7 years experience programming, so I'm all set on either side. Thanks for the offer though - good to know where resources lay (lie? never get that right - language is not my strong suit).

 

[jnuts]

As I have said before when it comes up, the neural nets tend to make more predictions, but aren't always right.

Have you ever tried to do any natural language processing to parse info as it comes in over the wire to modify your picks? Kinda like KEDS or TABARI does (different subject area, but similar concept).

http://www.ku.edu/~keds/software.html

 

[jnuts]

btw - why the name change? I'm not up on the latest happenins at EF.

 

[NoDaddyNo]

Jnuts, I changed my name because I like this one better. No other reason.

As for natural language processing - I have written lots of code to do such things - but have yet to direct it at financial data since it is hard to get enough historial data of news.
Back history of financial data is relatively easy to get a whole lot of.