Definitely highly optimized/curve fitted. Right now it seems the cost are roughly the same as the profit but I've estimated the cost very conservatively -> in reality it should be lower. System is ready for prime time, the risk allocation is 10k/position and unless I've messed it up it should generate trades starting tomorrow.
Thanks for that input. I've been struggling with that in the past quite a bit, always looking for a decent slow walking equity curve but you're absolutely correct, the smallest change messes things up quite a bit.
There is a metric on Multicharts that points out potential upside. I cant remember exactly what it is, but it makes sense. You need to be able to make big winners once in a while. A good high % system is not a bad thing, but they can under perform.
Read one more line .. Do you think a solid system works on most markets or do you think it's ok to cherry pick?
This is the heart breaker phase of system development, you turn it on, the initial live trades (4 so far) look much better than the backtest, you wonder if you were too conservative on slippage or commission and you think you've found gold (right now my average profit / trade is double what my backtest results estimated, the win ratio is in line). I've been here many times before and just like with women all of a sudden things change out of nowhere and go south . Just a matter of "is it today, tomorrow or next month" On a more serious note I realize four trades are not significant but the good news is that the mechanics are working correctly and now I just need to collect the results and see how they compare.
Comparing the averages gives you nothing. Look how far your live average/trade is from the backtest one, not on $ value, but in number of stdev - of course, you must use for stdev what it would be for your live trades sample size, not the backtest sample size. To get that stdev value, a quick & easy way is to run a MonteCarlo simulation using your backtest results as distribution, simulating a number of trades = your live trades.
dom, would I average the median/std dev then across the Monte Carlo simulations? Otherwise the variance would be way too wide to be meaningful since there aren't that many live trades yet but there are over 49.000 backtested ones. As I mentioned in the previous post, the median is 47.8, the std dev is ~ 480. (on a full position size [10k margin] based on the backtest results across the entire distribution, no monte carlo here). The live trades are half the position size [5k margin] the median is 45 and the std dev is 138. There's only 7 live trades as of now.