Bond Trader 2009

A cautionary note on neural networks

It has been my experience that looking for trading systems with neural networks can lead you down a time wasting path that you will regret. You will find that the network will do things that no one in their right mind would ever do. You'll find it trading too much. You'll find it trading too little.

I've spent thousands on software over the years. Probably 10's of thousands. I collect trading software like a kid collects baseball cards. Guess what I use to make my bread and butter trades? Amibroker. (I do use NinjaTrader for order entry.)

Yes, I have all kinds of NN software, but when I actually trade, Amibroker does everything I need.

Here's where NN software can definitely come in handy. Use it to fine tune some indicator that you think has merit. The use of this indicator is not to take signals from it. You use it to confirm direction.

The way I use my 4 min bar NN is to tell me that it is safe to buy/sell support resistance in the direction of the NN. I do not take signals from it. A second use of an NN study would be to determine the ideal times to buy or sell on any given day.

In retrospect, I look at the several years of precious time I misdirected into creating an NN system as time that would have better been spent on market geometry and price action.
 
My current workflow

Each day, prior to market open I do the following:

1) Glance at the daily and weekly pivots. I have these display on a 15 min chart that I keep minimized and only refer to on trade entry or exit.

2) Study the Market Profile chart. Really study it. Find the POC's for the last few days. Look at the ranges. Decide price targets.

3) Look at WGSR on the 60 min and the 5 min. The 60 minutes gives me longer term targets, for the entire day. The 5 min gives me targets that can occur within 15 minutes or so.

4) See if any of the above matches. Make a note of matching S/R. Usually this gives me only 2 - 4 immediate levels to be concerned with.

5) Look at my ideal times for the day. There are usually 2 to 4.

6) Look at my wave count on the 60 min and 15 min. The 15 min takes precedence over the 60 min.

The above takes maybe 15 minutes to 30 minutes. In a pinch I can throw out everything and only look at item 3 above.

After 0820 EST

1) Look at my 4 min momentum NN.

2) Be very cautious around important reports.

3) See what the opening hour range is.

4) See when an important time is coming up.

Trade execution

This requires patience. All of the above has given me a feel for where the market is going and what I an safely get away with in terms of entries and exits.

I do two things. I watch the hourly WGSR. It doesn't change often, and I know that I can safely buy/sell the WGSR for a minimum of one time. This gives me a single trade maybe once an hour.

Second, I watch the 5 min WGSR. I look to going in the direction of an established 5 min WGSR and will take an entry based off of a retracement.

That is it. With the hourly, I can enter the trade order an hour or more ahead of time. With the 5 min WGSR, I need to execute the trade real time. The hourly gives me possible profit targets of 8 ticks or more. The 5 min gives me a possible target of 4 ticks. Obviously, sometimes the trade continues in my favor and I hold on to the second contract.
 
0820 EST

I'm looking to short 116 080 at the moment, for a first approach only.

If we get weak, I may jump the gun and shoot for a scalp short entering at 116 030 and exiting on a dip below 116 000.

After that trade, I'll be looking to short 116 140 to 116 160 for a first approach only.
 
0826 EST

We are looking just a little stronger now, so I will sit tight and wait for my price of 116 080, as stated.

Today may be a bit slow so not trading is OK too.
 
Quote from johnpinochet:

A cautionary note on neural networks

It has been my experience that looking for trading systems with neural networks can lead you down a time wasting path that you will regret. You will find that the network will do things that no one in their right mind would ever do. You'll find it trading too much. You'll find it trading too little.

I've spent thousands on software over the years. Probably 10's of thousands. I collect trading software like a kid collects baseball cards. Guess what I use to make my bread and butter trades? Amibroker. (I do use NinjaTrader for order entry.)

Yes, I have all kinds of NN software, but when I actually trade, Amibroker does everything I need.

Here's where NN software can definitely come in handy. Use it to fine tune some indicator that you think has merit. The use of this indicator is not to take signals from it. You use it to confirm direction.

The way I use my 4 min bar NN is to tell me that it is safe to buy/sell support resistance in the direction of the NN. I do not take signals from it. A second use of an NN study would be to determine the ideal times to buy or sell on any given day.

In retrospect, I look at the several years of precious time I misdirected into creating an NN system as time that would have better been spent on market geometry and price action.
More...

I have noticed that much time is wasted in NN by many (too hard to troubleshoot when things go poorly).....have you tried out any genetic algos?
 
0834 EST

Ideal trade times today are provided as a range:

0900 to 0930

and

1130 to 1230

It is the best I can do at the moment. More pressing concerns.
 
0903 EST

The trade is taking a lot longer than I would like, however that is the risk one takes trading a slow day. I'll look to exit before 1000 EST to be safe, unless we are on a roll.

I'll be watching this trade closely, and I might even exit with a quick breakeven trade if we reject 116 060 strongly again.
 
0937 EST

Probably time to get out now if you are still in. Exit would be 116 050 or better.

My wild guess is that we drift lower to 115 310 or lower BUT the 1000 EST report is coming and on a slow day it may have some impact. Better to lock in a profit on a trade that almost hit my stop.
 
You must log in or register to reply here.
Back
Top