Looks like some funds really wanted to sell early this morning. I've been watching the PREM get so low the buy index arb. programs were buying while the market was dropping. Somebody must really want to sell if they're that desperate. It'll be interesting to see if the market makes up much ground when the selling wave stops.
10/15/03
- Thread starter acrary
- Start date
What kind of trades did you do this morning, acrary? As I recall you like to make your first trade in the same direction as the opening gap. Is that what you did?
By the way, I came across your statistical validation post (look at past occurences of your pattern and compare the moves to random moves of the same duration) and found it very educational. Where'd you learn your statistics? Do you have an advanced degree? Thank you for your posts.
By the way, I came across your statistical validation post (look at past occurences of your pattern and compare the moves to random moves of the same duration) and found it very educational. Where'd you learn your statistics? Do you have an advanced degree? Thank you for your posts.
My models are all breakout driven. They give me points where I need to be long or short for a day.
Today I'm a buyer. I've been taking my first two trades each day by frontrunning the index arb programs. It looks like free money early in the day. Just blew my mind sellers would dump stock without selling futures first to lock in their price. My models think anything above 1053.80 in SP is likely to continue up today. I almost got trapped near the open. My original plan was to buy on a successful retest of the close or on a breakout above 1054.00.
Any more breaks below the low and I'm done for the day.
I'm a stats novice. I have a friend who has a PHD in mathematics from MIT who has helped me come up with techniques for doing analysis. I trade his account for him as payback for his help. Same setup I have with another friend that is a neural net expert.
We'll see if yesterday's close becomes resistance. If not, I'll probably start doing some serious buying for the day.
This morning's high (in the day session) was exactly 1053.80. Just wondering, Did your model give you this number based on EOD up until last night?
Have you actually found any positive results using neural nets? (From my limited knowledge of the subject, there appears to be some recent promising results using SOMs (Self Organizing Maps)).
Richard
Well, I just shorted Krispy Kreme (KKD). I figure it is bearish when the market is "taking profits" on all the good news. Plus we are overbought and the VIX is pretty low. And I'm shorting KKD instead of QQQ's because I think KKD is overvalued relative to the market.
I don't share acrary's bullish bias and I hope I don't regret it.
I plan on holding for 3+ days.
I don't share acrary's bullish bias and I hope I don't regret it.
I plan on holding for 3+ days.The model derived it based on the EOD SP day session data.
Another one will buy if the market hits 1055.10 today.
Have you actually found any positive results using neural nets? (From my limited knowledge of the subject, there appears to be some recent promising results using SOMs (Self Organizing Maps)).
Richard [/QUOTE]
Thanks for the SOM idea. I have it available in one package I own, though I've never looked at it. I'll investigate it further.
Here's what I told another person in a PM about how I've used NN's.
I've used Brainmaker Pro with the Genetic Training (GA) option.
http://www.calsci.com/products.html
Here's a example of how I used it:
Let's say I want to see if I can find patterns to predict buy days when the low is made in the first two hours and the high is made near the close. I would write a program to find all days within a year that had that happen. Then I would use the daily data for say the previous 20 days and lots of indicators and load it into the NN package. If I tried to have the NN train on all the data it would weight all inputs and I'd have a mess. Instead, by using the GA to weed out the inputs, the training goes better and I'm left with a smaller subset of variables. I'd then use those variables with another NN package (NeuralWorks Professional II/Plus) http://www.neuralware.com/products_pro2.jsp to find the weights of each variable that did well. Then I'd spend time looking at the heavier weights and try to develop a set of patterns that looked promising. I'd then do the tests on each pattern. I'd add those that looked most promising (70% mean with a standard deviation less than 20) to a test program in Tradestation. Once I had a decent sized collection, I'd start doing the year by year testing and check for a consistent edge.