Following or Predicting

Following or Prediction?

Prediction.

I’ve developed a neural network that tries to predict moves. It’s uses proprietary indicators as inputs. The system was trained on about 2 dozen stocks, optimizing only 4 parameters (the same parameters value are used for all stocks).

Here’s a 1-year out of sample equity curves for individual stocks.
 

Attachments

TA should equate to an instrument in your car. It tells you what is happening. It is objective, if you use the right methods.

If the instrument tells you something bad is happening, then you have to take action or your car will stop running. {If you don't then you are a loser] If the instrument tells you everything is all right, then just let the damn thing run.

I wouldn't speculate about how my car will run 15 minutes from now, or 2 days from now, or 2 years from now. I just watch what is happening and make the adjustments along the way.

Any other use is just plain BS.
 
Quote from WolfVector:

And here’s the out of sample equity for an equally weighted portfolio of the same stocks as above.
More...

"NN" as in "Nearest Neighbour"?
 
Let's keep an objective if not scientific mind about this and all matters.

Common view: TA can't predict and doesn't work.

Commissioner of Patents in 1899 - "We should close the Patent Office, because everything that could have been invented, has been invented."

Informed medical opinion pre Bannister: Running a 4-minute mile is impossible. After he did this feat he was considered a freak of nature.

Scientists held this opinion: The human body cannot endure the stresses of travelling above 30 mph.

Fact: Never underestimate the human capacity for ignorance. My motto in TA is "Everything Works!" So the question simply is, am I good enough to make it work?

The good book says, "As a man thinks in his heart, so he is."

My personal discoveries are man will never stop inventing, discovering, creating, improving, making hitherto impossible breakthroughs etc.

TA predicts. Moving Averages don't lag - they can predict. TA can follow - ie keep you in a trend. It works for prediction and following down to a sub-1 minute level.

To discover these facts about TA all you have to do is pay an immense price in hard graft. You won't find it in books.

Good trading
 
Being objective is the key. It’s just that most people think that something like buying the breakout of a down trend line is being objective. When in fact, they don’t know how often buying a breakout of a trend line works nor do they have an objective way for plotting the trend line. They don’t have an objective way to determine when to take profits.

All of these issues can be solved by taking a scientific view of a technical setup for taking a trade and performing basic research. Can the setup be identified with no judgment? How often does the setup occur? What is the range of outcomes? Is there an objective exit point? You can be objective once this research is complete and you have statistics to analyze. It is a large amount of work.
 
Quote from WolfVector:

And here’s the out of sample equity for an equally weighted portfolio of the same stocks as above.
More...
Slippage? Transaction fees? Buy-and-hold index?
 
kut2k2: “Slippage? Transaction fees? Buy-and-hold index?”

Commission and slippage were included in the test. QQQQ, SPY and IWM were included in the portfolio and had the following performance. Note: Annual Percent Change in Price for the ETFs below is equivalent to buy and hold without accounting for commission and slippage.

OUT OF SAMPLE TEST
Start Date 9/23/2005
End Date 9/22/2006


Nasdaq 100 Index (QQQQ)
---------------------------------
Beginning Price 38.75
Ending Price 39.87
Percent Change in Price 2.9%

Out of Sample Test Results:
Annual Return on Trades 53.9%



SPDRs S&P 500 Tr (SPY)
-------------------------------
Beginning Price 121.44
Ending Price 131.47
Percent Change in Price 8.3%
Annual Percent Change in Price 8.3%

OOS Test Results:
Annual Return on Trades 36.0%



iShares Russell (IWM)
----------------------------
Beginning Price 65.18
Ending Price 71.54
Annual Percent Change in Price 9.8%

OOS Test Results:
Annual Return on Trades 55.5%
 
Hello WolfVector,

I am currently looking into writing a neural network program for short term trades. I understand that you're keeping everything secret but I am curious about a couple of things if you don't mind sharing. I'm trying to get an idea of what I can expect as a reasonable goal for the possible performance of a net.

1) What kind of capital are you using to get these returns?

2) How many positions at a time would be open?

3) Are you long and short or just long or just short?

4) When you are saying out of sample - are you referring to the time frame or to stocks that were not used in training (I was thinking of doing the latter)?

Thanks for whatever you are willing to share.
 
You must log in or register to reply here.
Back
Top