What you're doing is roughly equivalent to buying a synthetic put option on your own performance. Such an option can't be costless, although it might be in a particular simulation just by fluke. I've also done work on this kind of thing. I think its reasonable to "buy" a massively OTM option which will cap your downside at the cost of a tiny bit of sharpe, just don't expect it to have zero cost. For example you might want to "buy" an option which starts scaling down your risk when you get close to the maximum drawdown you see in simulation.
Not sure if I'm following you here completely...hedging with options here would be rather tricky as I'm shifting between long and short futures positions on a daily basis...?
My unsolicited opinion below: Your PNL swings are too big. If a 30% drawdown is expected, what will happen to your risk tolerance and margin capacity afterwards? That can have a drastic impact on your dollar performance (which is the space high water marks are defined in). If you are seeking insititutional capital, delever your returns so that it becomes a +25% return and -6% drawdown. If you draw down 6%, you will be able to continue trading at a comparable after the drawdown.
Wouldn't want to put words in GAT's mouth but what he is saying is that by trying to time your equity curve you have basically paid to overlay a de-facto put option on top of your original strategy. I have seen such overlays being done by management on systematic strategies but its more of an art than a science.
NMM is exactly right - I'm talking about a synthetic put not a real one. "Pricing" these puts (seeing the effect on your p&l) is difficult. You can do it with simulation, but if it only comes in at very deep drawdowns then its a matter of luck what the "cost" is. Its better to use monte carlo with repeated draws from your return distribution (which is going to be equivalent to a non parametric option pricer); or you could just plug in your performance stats to a closed form option pricer (though for very deep drawdowns this will be an approximation as these tend to be brought about by a combination of time series dependence and skew, which vanilla models don't handle).
Guess you are right here, however, this client (who happens to pool investment from 3,000 HNWIs) explicitly wants this rate of return, what's better to do, deleverage or give the client what he wants?
Well, currently I actually apply this in two steps (i.e. three levels in total). At the first level, the equity curve is like a roller coaster, at the second level it becomes a kind of upward sloping staircase function (but with the recent drawdown), and at the third level it's identical to the 2nd, but avoidance of the recent drawdown (instead performance is flat during that period). I've been experimenting with having a 2nd algo which is trading when the original version is inactive, however with rather limited results as the choppiness is really hard to cope with...
Btw, I agree that money management has to be a lot tighter - as I haven't experience a drawdown of this magnitude I was simply too optimistic for a turnaround at first, instead I should have halted order execution much earlier, at $1,000 below the peak on the equity curve - simulations show that this would have maximized total return. Alas, this means more passiveness though, which in the long run calls for deployment of alternative algos to take its place on instances like these...
Have you tried comparing your strategies on other futures instruments / different timeframes and using different chart resolutions? You may find that performance has improved on another instrument and maybe you can 'hedge' by running the strategy on an instrument that is responsive to your algorithm.
Haven't tried other similar instruments like Russell 2000 etc., but in my experience chances are that it will work even worse as even the slightest difference in characteristics often has a negative impact, anyway might be worth trying...