Long time member but more of a reader than a poster :)
I'm using CQG- IC for my platform and almost exclusively trade futures.
Curious on what metrics others would focus on during backtesting. My options are as follows:
Net Profit & Net Closed Profit
Average win per trade-- (a average off all trades (winners and losers) giving a net expected (avg) return on any given trade.
Profit to Max Draw-- Net profit as a ratio of the maximum draw down
Profit/Loss Ratio
Percentage winners
Remove to Neutral -- a percentage of the total number of trades (starting w/ largest winners in descending order) that could removed before the strategy profit would drop to a break even.
Return Retrace -- Average compounded return divided by the average maximum retracement (AMR), where the AMR equals the average of the maximum retracement for each point and the maximum retracement equals the larger of the: maximum retracement from a prior equity peak or the maximum retracement to a subsequent low. (I rarely turn this on as it adds time to the backtesting)
Linear Regression --Applies a least squareslinear regression line to the profit curve producing an average return,which is divided by the Standard Error. (Also generally leave this off due to time to Backtest)
There's other ( what I feel are typically less important) metrics like "Time in position", Max consecutive losers (& winners), Max Drawdown Duration, largest loser/winner, largest closed drawdown, avg loser, avg winner (not the same as Average Win per trade)
Typically my process is to prioritize "Profit/Max Draw" and then look at the range of returns, picking out a balance between this and "Remove to Neutral" as well as "Average win per trade" and of course "Net Profit". I've never figured out a way to rank these or prioritize them one over the other. I also keep an eye on the equity curve and try to concentrate more so on curves that are 'smoother' visually. I've worked through this process using backtesting on a large range of data (6-10yrs) and then carry it forward with a segment of data that is not backtested (usually 1-2yrs), followed by walk forward and then ultimately "real-time data" testing using CQG's Autotrading systems. To date I've not found a system that will produce constructive returns over the long term. I've had systems that were wildly successful to which I switched to real money after a time. However, after 12-18 months they will fail equally as spectacularly.
Curious what others think of some of these metrics as a priority or my method of testing.