My strategy in summary: slow trend following, diversified over 40 markets, about one trade per market per year on average (3/4 of the trades' duration between 3 and 11 months), very wide initial stop, position size is determined by ATR at the time of entry (no volatility targeting afterwards).
I did not perform any curve fitting. I simply picked the parameters that approximately resulted in certain trading speed I had in mind. Then I checked neighbouring parameter values to make sure it doesn't fall apart. Then out-of-sample tested it on another portfolio. So I believe the system is quite robust.
That isn't a million miles away from what I do. I trade a bit quicker, but over the same range of futures markets (I'm assuming you trade futures).
I don't use explicit stops, and I do continous trading rather than discrete (so I'm always adjusting positions according to changes in vol and forecasts). Based on research I've done for my most recent book in expectation that will add about 40% to performance. So for example if your Sharpe Ratio was 0.9, then it would go up to 0.9*1.4 = 1.27.
Is it worth doing this? Well it's a fair bit more work if you're not fully automated. So it depends on how much capital you have, and what you value your time at.
As I understand you do a sort of bootstrapping test by selecting random portfolios from the pool of instruments that you have in your database and come up with confidence intervals for certain performance figures (CAGR, Sharpe, max drawdown, ...)?
Ok so for each instrument I have a set of trading rules, each with variations of different parameter sets, in total perhaps 40 different trading rules. So for example a moving average crossover could be 4,16; 8,32 and so on. The set of variations is selected using artifical data, with no reference to performance. I drop trading rules that are too expensive for a given instrument. Then I use a portfolio selection method to decide which combination of rules to trade, and how much weight to give each one. Although I've written about bootstrapping, my preferred method is actually something called 'handcrafting' (more details
here).
Importantly, I will trade all the rules that are cheap enough to trade for a given instrument, I don't just pick one.
Then I again use a portfolio selection method to decide how much risk capital to allocate to each instrument I trade. It's likely that the result isn't that different from what you would end up with if you started with the same rules, as I'm going to drop the most expensive rules first, but:
- as it's an automated selection process it can be backtested.
- I am trading multiple parameters for a given trading rule, it sounds like you only trade one. In expectation this will reduce your performance.
- The combination of trading rules I use determines the holding period, whereas for you the ratio of ATR to stop loss will determine your holding period. So there is never a potential mismatch in speed between entry and exit rules.
The final source of extra performance is speed; you are trading very slowly, and although I generally advise people to slow down, there is likely to be some benefit from including some faster trading rules if you are only doing one trade per year (this also means your performance is going to be very similar to the underlying markets, so your 'information ratio' won't be very high). This is especially true if you are only using one rule per instrument, so you can get a diversification benefit from adding more.
GAT
