Welcome to NexusFi: the best trading community on the planet, with over 150,000 members Sign Up Now for Free
Genuine reviews from real traders, not fake reviews from stealth vendors
Quality education from leading professional traders
We are a friendly, helpful, and positive community
We do not tolerate rude behavior, trolling, or vendors advertising in posts
We are here to help, just let us know what you need
You'll need to register in order to view the content of the threads and start contributing to our community. It's free for basic access, or support us by becoming an Elite Member -- see if you qualify for a discount below.
-- Big Mike, Site Administrator
(If you already have an account, login at the top of the page)
Hi Bizman - it's tough to nail down the exact ratio or even the data range period. We attempt to select an optimization range(training range) which is representative of all types of market conditions leaving the balance of the data range to feed forward(validation). Please note that we consider CH results as preliminary; we further evaluate the model from with TS/MC & NT.
1 - Data Range 30May10 - 7Nov10 Training = 64% / Validation = 36%
2 - User can select other ranges
3&4 - CH in training, note 8 processors sharing the load;
I've read at least a hundred articles on neural networks and everyone has an opinion on what type of data to feed the model; pre-processing; post processing; etc ... it's more of an art than a science, so it seems.
While we haven't even scratched the surface on model building we have some good ideas on employing and monitoring model performance.
///tortexal - try the NN_TRIX strategy first. We modified the NWA_ES to clamp the predicted output. We have not posted the revised code due to lack of interest.
can you post you strat inputs for this time period? this is for the NWA_ES_R3a system. I ran an optimization from 9/26 to now, then looked at performance from 1/1/2008 and its straight off a cliff. i probably do not have the right inputs
Here's performance w slippage going back to 1/1/2008. As you can see, the use of mkt orders and not accounting for even 1 tick in slippage completely destroys the strat, which is an enormous problem. a non HFT robust system should take into consideration mkt ineficiencies when filling orders. A recomendation here is to reoptimize your sample period with slippage included. as someone who runs client money on automated systems, i have never seen a market order fill without incurring slippage in any market in any vehicle. 100% chance you will have slippage in order mkt orders. that being said it is unreasonable to not control the controlables when building a system.
Then to refute the previous statement that stops and profit targets only hurts performance, below are the exact same settings as above but with basic money management:
As you can see, you go from losing $38,000 to losing roughly 300 bucks. Quite the improvement even on a poor performing system
When planning a trip to vegas, it is best to have some kind of strategy in sofar as when to get up and leave the hold em table. I leave for two reasons. 1: to keep my winnings 2: to prevent me from going on tilt
tables have a min buy in amount. many people use the rule of thumb of having a minimum of X # of big blinds in chips. When they drop below X, they leave. The inverse is that if im able to chip up X% of my buy in, i also leave.
mkts are no different, especially when it comes to auto trading systems. So the reason the stats reflect taking less trades is because im simply leaving the table after running out of chips, or making a good amount vs my buy in.