Here's the net profit chart to parameter value. Some pretty noticeable improvements over the take any trade strategy.
NPP builds a Emini system
- Thread starter no_pm_please
- Start date
so I really ought to have tested this myself before blabbing away! amgI changed my backtest to 2001 - Oct. 2006 to see if it showed any curve fit tendencies. Here's the annual summaries with the 2 additional years. The profits in 2003 got a boost because the 100 bar exclusion in TS wasn't affecting it anymore.
omniscient
Guest
not sure if this is what he meant, but i read it as:
if the current 14 period ADX is less than 101% of the previous 14 period ADX ...
hth
take care
omni
I use adx often as a tool to measure trend. It's a simple tool. Others many want to use pivot points, etc. One of it's flaws is you'll also enter reversal trades on the pullbacks, which probably adds some losing trades.
In this case a falling adx with a crossover through a midpoint rsi value can only happen if there was a trend and a recent pullback. I added the 1.01 to indicate a flat adx as well which can happen if a trend flattens out and the rsi midpoint moves to where the flattening happened.
Just using simple logic. Checked to see if locking in a larger profit instead of allowing larger volatility was the way to go. It was.
Also checked to see if condition of trend has any influence on the profitability of using a general trend following approach. I checked a rising adx which could only happen from either a dead market or a complete reversal of the previous trend. That was a poor strategy. Next, I checked to see if level of adx had any influence. I tried using low adx numbers versus high adx numbers. It didn't affect the profitability at all. The last check I did was to see if pullbacks within a trend had a improvement. It was significantly better, so that's the general approach I'm taking. It also appears to be stable over time.
I calculated the expected return for 2003 - 2006 using the different approaches to see if a edge was present. Using a 1% risk per-trade with no compounding here's the results.
Original starting method
2003 -7.89%
2004 42.67%
2005 10.7%
2006 -9.77% (through October)
Total expected return 35.71%
Using optimization of 2003 - 2006 the total expected return drops to 25.98%, so choosing a single general purpose number for the variable is a poor strategy.
With pullback added
2003 38.58%
2004 6.1%
2005 26.69%
2006 18.49%
Total expected return 89.86%
Using optimization of 2003 - 2006 the total expected return drops to 72.77%. so again, using a single number for the parameter for the entire time is a worse strategy than a random approach.
Using the 84 parameter
2003 82.42%
2004 20.16%
2005 41.9%
2006 66.6%
Total expected return 211.08%
Since I have no reason to believe the 84 number is any more likely to be the correct number going forward, a better approach is to look for a better parameter selection strategy than using all the data and doing optimization. I'll try to come up with some ideas that can beat the random parameter selection to improve the edge.
The nice thing is this idea has a positive edge as long as the parameter ranges between 10 and 99. Although the best parameter could grow to 100+, I think it's pretty unlikely.
Original starting method
2003 -7.89%
2004 42.67%
2005 10.7%
2006 -9.77% (through October)
Total expected return 35.71%
Using optimization of 2003 - 2006 the total expected return drops to 25.98%, so choosing a single general purpose number for the variable is a poor strategy.
With pullback added
2003 38.58%
2004 6.1%
2005 26.69%
2006 18.49%
Total expected return 89.86%
Using optimization of 2003 - 2006 the total expected return drops to 72.77%. so again, using a single number for the parameter for the entire time is a worse strategy than a random approach.
Using the 84 parameter
2003 82.42%
2004 20.16%
2005 41.9%
2006 66.6%
Total expected return 211.08%
Since I have no reason to believe the 84 number is any more likely to be the correct number going forward, a better approach is to look for a better parameter selection strategy than using all the data and doing optimization. I'll try to come up with some ideas that can beat the random parameter selection to improve the edge.
The nice thing is this idea has a positive edge as long as the parameter ranges between 10 and 99. Although the best parameter could grow to 100+, I think it's pretty unlikely.