Just show me one mechanical technical analysis that has been profitably backtested?

Quote from Hook N. Sinker:

These are the results of trading simulation using 46.83 years of General Motors Corporation daily closing price data. Heat is 5 %, initial capital $ 100,000, assuming $ 10 commission per trade.

Buy when price value increases 10%
Sell when price value decreases 10%
Total Profit $ 54088

Buy when price value increases 15%
Sell when price value decreases 15%
Total Profit $ 228424

Buy when price value increases 20%
Sell when price value decreases 20%
Total Profit $ 217186

Buy when price value increases 25%
Sell when price value decreases 25%
Total Profit $ 83722

Buy when price value increases 30%
Sell when price value decreases 30%
Total Profit $ 27906

Buy when price value increases 35%
Sell when price value decreases 35%
Total Profit $ 29628

Buy when price value increases 40%
Sell when price value decreases 40%
Total Profit $ 91331

Buy when price value increases 45%
Sell when price value decreases 45%
Total Profit $ 204733

Buy when price value increases 50%
Sell when price value decreases 50%
Total Profit $ 148791

===

I find lots of methods show profitable simulation results. I also recall hearing stories about people losing money in the stock market. If it is not the mechanical system that loses money then the problem might be in the human part of the system. People might not follow their own rules.
More...

What a horrible system. 48.63 years for a 50% return? That sucks. Even a 200% return is nothing to write home about in 50 years.
 
245hjj7.png


Including 2% commission per trade and 3.5% slippage. Historical tick for backtest, but uses BBO calc. for realtime trading to minimize the slippage.
 
If your definition of "mechanical technical analysis" includes OHLC formations then this program will find you hundreds, if not thousands of systems, across several markets:

http://www.tradingpatterns.com/

Of course, the problem is whether these systems will stay profitable in the future.

Any ideas for a good test other than a walk forward?

I think the key is the "test", not the system. Even if the system has been profitably to this date, it can turn to a loser at any time.
 
Quote from intradaybill:

If your definition of "mechanical technical analysis" includes OHLC formations then this program will find you hundreds, if not thousands of systems, across several markets:

http://www.tradingpatterns.com/

Of course, the problem is whether these systems will stay profitable in the future.

Any ideas for a good test other than a walk forward?

I think the key is the "test", not the system. Even if the system has been profitably to this date, it can turn to a loser at any time.
More...

It depends on how you back test the system.

Most people take a few thousand or tens of thousands of bars, develop methods and strategies to trade and then test on say the next 10,000 bars (some period that is a significant portion of available data but much less that the development set. Testing at the end of the data has a lot of drawbacks in that it assumes this period is typical and it will continue during live trading.

A much more effective approach for creating systems that are stable and robust is to break all past data into two sets: one for system development and the other for back testing after using an unsupervised SOM to assign a cluster identity to each bar.
By splitting the set first by the bar identity and then into development and back test you insure that there is a proportionate distribution of all market conditions drawn from the entire set of available data for both development and testing.


Jerry
 
Quote from Jerry030:

It depends on how you back test the system.

Most people take a few thousand or tens of thousands of bars, develop methods and strategies to trade and then test on say the next 10,000 bars (some period that is a significant portion of available data but much less that the development set. Testing at the end of the data has a lot of drawbacks in that it assumes this period is typical and it will continue during live trading.

A much more effective approach for creating systems that are stable and robust is to break all past data into two sets: one for system development and the other for back testing after using an unsupervised SOM to assign a cluster identity to each bar.
By splitting the set first by the bar identity and then into development and back test you insure that there is a proportionate distribution of all market conditions drawn from the entire set of available data for both development and testing.


Jerry
More...

An additional note: the results generally increase the ratio of profitable trades to loosing ones by 8 to 20% depending on the market and bar lenght of the trading system. It works better on Forex than Futures for some reason.
 
Quote from Jerry030:

A much more effective approach for creating systems that are stable and robust is to break all past data into two sets: one for system development and the other for back testing after using an unsupervised SOM to assign a cluster identity to each bar.
By splitting the set first by the bar identity and then into development and back test you insure that there is a proportionate distribution of all market conditions drawn from the entire set of available data for both development and testing.
More...

Hi Jerry,

Do you know of a program that can do this as part of the back-testing analysis?

By SOM did you mean "self-organizing map"? I have tried something along these lines using NN before but it did not work very well.

Best wishes for the Holiday Season!

Ron
 
Sure...I have been using this daily swing system for many years: [/QUOTE]
----------------------------------------------------------------------

too good to be believable - possibly.

Apart from this - basically a wonderful thread but too few details.
However, some wonderful inconspicuous self-portrayals. lol
 
You must log in or register to reply here.
Back
Top