The attachment is an image of the state of the swarm as at the last bar. The bots that are spread throughout the entire parameter space are 'ghost' bots. Ghost bot signals are not used when generating the systems trading signal, but they are used during the bar by bar adaptation process. The cluster of bots is the 'live' swarm.
The attachment is the trade report for the S&P 500 ARM03 system using detrended data. As I mentioned earlier, the trade report needs some work.
I used the following parameter ranges for the QQQQ SMA Crossover system: Plus SMA length 2 to 8 Minus SMA length 20 to 40 The lookback period is 1,000 trading days. The results are not anything special. The attached file is an image of the equity curve.
The results you get are very disappointing. This is what I get with a stop and reverse system that buys when mov. avg 5 > mov. avg. 30 and sells when it is <. Equity curve is like a straight line up. Total net profit 88.2400 Total long trades 333 Total short trades 49 Winning long trades 236 Winning short trades 11 Total winning trades 247 Total losing trades 135 Win rate 64.65% Profit factor: 1.85 Max. open drawdown -3.30 Obviously, your bots are doing something wrong. I get the feeling from looking at your open equity of about 8 versus the closed equity of just around 6.6 and the low win rate you get that the bots are taking the extreme parameter values.
Thanks for the trade report. I don't think it is the bots. I am using SMAs of the close and trading on the next open. Are you doing the same?
Yes, I use that file goodgoing uploaded. Trading starts on Jan. 2002. Actually, it turns out that the (5,30) system is a good one. I wonder why your bots fail.
It is related to the trading delay. I just set the trading delay to zero and the equity curve looks good. However, my understand is that a trading delay of zero means the system will trade on the current period's open. I need to take a close look at the trades. I don't usually trade on the open so I might have used the wrong setting.
According to the Dakota documentation my initial understanding of how the trading delay works was correct. I am going to check all values on a spreadsheet against a 'fixed' 5/30 parameter system in Dakota.
I checked a Dakota 5/30 system against a spreadsheet and they were in perfect agreement. I am using simple moving averages of the closing price and trading on the next open. The equity curve of the fixed 5/30 system is uglier than the adapted version.