Silicon Investor (SI) -- The First Internet Community

STOCKTALK

We've detected that you're using an ad content blocking browser plug-in or feature. Ads provide a critical source of revenue to the continued operation of Silicon Investor. We ask that you disable ad blocking while on Silicon Investor in the best interests of our community. If you are not using an ad blocker but are still receiving this message, make sure your browser's tracking protection is set to the 'standard' level.

Strategies & Market Trends : Neural Nets - A tool for the 90's -- Ignore unavailable to you. Want to Upgrade?

To: LastShadow who wrote (743)	4/5/2000 10:26:00 PM
From: xu, b.	Read Replies (2) \| Respond to of 871

Message #743 from LastShadow at Apr 5 2000 8:49PM I would suggest the following: Tech stocks from several industries - 10 or 12 stocks Bai: agree. Just a thought: would it be better if we could choose stocks from semiconductor sector to train the nets and test them on bio? (Long-term fundamentals suggest bio-party is just beginning especially in bio/semi marriage areas.) If the nets don't fall apart this way, it's robust enough to trade. compare to the sector spyders or industry indices Bai: agree. We need to group stocks in the same sector in order to compare to sector spyder. Or am I missing sth.? I prefer training data sets of 2-3 months, as amount of time data isn't as beneficial as different types of data (characteristics) Bai: both you and Dan seem to favour short training sets. My intuition is that kind of models need frequent retraining. But I will follow you guys. It may make sense as institution dominated market moves really fast. I have a feeling that market cycles become shorter and shorter. maximize profit - manage risk with the trading strategy we will evolve once we understand and coorelate the nets output to tradeability (stress limit entries/exits, with appropriate stops, which can also be modeled) Bai: agree. The criteria I use to evaluate systems is P/R as I explained. set a fixed number of shares (min 300, max 1000 as that minimum is the lowest number you can use for all-or-none trades) this will make computation easier and allow mangement via a position trading account guideline - I will post those later, or you can research my posts from a couple years ago. Bai: From system comparison point of view, I favor equal dollar amount if this is possible with your s/w. Equal share approach has a bias towards high priced stocks. Your point regarding trade execution is well taken. I use Datek and All-or-none order is not an issue. For discussion purpose: I droped all-or-none order quite a while ago. If one's orders frequently got partial fill, it's an indication of liquidity problem, meaning one's style/approach often requires fill at extreme prices (H/L), where the market is very thinnly traded. It's kind of winning systems that are not tradeable especially when you wnat to move more than several thousands of shares around. (I am not there yet.) Another possibility of partial fill is fast market move. In that case, partial fill is still good than nothing no matter the market is with you or against you. All-or-none order was useful when commission/trading asset ratio was an issue. I am probably off topics because trade execution is very important but we may not want to discuss it in this thread. I don't mind exchange more ideas on that issue in private. And we are interested in some sorts of template/generic models as you suggested, right? I recommend not looking for stocks that may pop up the next couple of days, but rather a robust best entry, optimim exit method - stocks that pop, in my experience, also are found with that method. We want consistency first, and one can then evolve that later to suit your own personal style/objectives/predispositions. Bai: agree with your philosophy. But I may not understand this "pop up" you suggested not to look at. It seems to be a good thing to look at as an investor/trader's job is to maximize return in the shortest timeframe possible. What I missed? I'd like to think we can create a model, a template for others to start with, and for us to build on. A learning tool for others, and a collaboration for us. Bai: agree wholeheartly. I always value a collective effort.