Backtesting SR is about 2.0 in backtesting which was done for a total of 5 years, this system only trades S&P500 e-minis. The degradation is clearly a result of different market regime so not a result from slippage. After having show such consistent performance for more than 2.5 years (equity curve looks almost like a staircase function), it appears very unlikely to be just a stroke of luck. NO optimization has been needed so far, all my trading systems are designed to work without parameter reconfiguration. Only change now is that I've implemented a order execution "circuit breaker" to prevent further decline before starting to see performance improvement.
It seems most likely the world has changed then. Trading this quickly its very likely you're picking up on market patterns that don't last for a decade. You're going to need to a new system, and a process for finding a new one that will work for a couple of years before you most probably need to find another one. If I might make a suggestion if you're looking at institutional money then they'd probably be uncomfortable with something of this fragility. I would work on a) a repeatable process for finding the new system, which you could demonstrate that it was robust by backtesting it (the process, not just the new system) b) Relying on a single market would personally terrify me. If you had more markets then you could more easily ride out these periods of poor performance giving some breathing space and making the portfolio as a whole more robust.
Good points, thanks! I have several systems, however they are sort of designed with different objectives in mind, e.g. some are aiming at absolute return, some relative, some maximum risk-adjusted return etc., hence not really interchangeable. Multiple systems with same objective is in my plan, however simply haven't had time/resources so far to achieve a complete "arsenal" of strategies... :-/
emini S&Ps only...few hours hold time...not optimistic about this. What effect are you trying to capture? It seems at first blush that this algorithm is only going to work in certain conditions and you don't have enough information to accurately judge whether these conditions are likely to be in force. Therefore you are left with the limits of backtesting and the accompanying uncertainty during a draw down (or will it be a blow up?). It seems you don't know the cause of the effect you are capturing and when/if it is likely to return. Not quite, but close enough. I think you've inadvertently stumbled upon something which benefits from a one of -a participant time slicing a large order. Likely long side. -the absence of aggressive liquidity consumption opposite your trade There is little value in this particular class of participant taking new long positions in this part of the market cycle. And the absence of other participants / strategies results in lower liquidity markets with more intraday volatility. So the effect you were capturing no longer exists and you're out transaction costs and whatever negative alpha your position management techniques create in the absence of the initial effect. What you are now looking for is a signal to tell you when to activate your algorithm and when to retire it. Depending on the size you are able to run it might well be worth bringing someone in to help you develop this signal. Perhaps put feelers out for execution specialists in this product. There will be people who understand enough about the product to make an educated guess at what your edge might have been and why it is no longer effective. Very few. And it would need to be worth their time. As an alternative you could look for a sort of natural hedge - for example you might find that your strategy does best in times of low volatility, and that being long vol will reduce your return in good periods but quid pro quo will offset your losses during a drawdown. This might increase your overall sharpe.
In principle I think you are right, but isn't that what everyone tries to do, i.e. identify and exploit different market regimes? By the way, we are already monitoring day-to-day performance so as to identify and differentiate between "favorable" and "unfavorable" trading days - it's been working surprisingly well so far, however, as futures trading is extremely unforgiving, the recent string of losses have virtually pulled the rug under our feet...
Sounds like you have some kind of a long bias AND long vol/volume in this setup? At least thats what I would guess based on the info. Would you be able to prove to the investor such is not the case? Because if so, I can't imagine an institutional investor getting into such a profile at this stage of the game. In any case, as others have suggested the move is probably to cut capital committed and keep going and I suppose you already are doing that by being soft kelly.
I wouldn't say there's a long bias by design, but since markets on average tend to rise in the long run, the long bias will follow accordingly. However, the system takes both long and short positions, hence in markets of decline it's supposed o benefit equally. Furthermore, it tends to perform better in periods where VIX is above 20, as it then tends to be more active, however low intraday volatility is preferred so I cannot say it is really long vol either...why wouldn't institutional investors be interested in the setup you describe above?
You are asking yourself a question I have also asked myself many times. Is this the drawdown that ends it all or just another run of the mill drawdown. Even when you carefully avoid look ahead, suvivorship, data mining or the other miriad research biases you can delude youself with to create misplaced expectations, the fact remains that the highest drawdown for a perfectly fine working strategy is still in the future. I have come to accept that all strategies I trade will work until they don't. Without further insight into the "reason" as to why the strategy should pay, and wether the "reason" is still valid, it will be very hard to tell what kind of drawdown you are in by just looking at the equity curve. This is the reason I find Kelly or fixed fractional type risk allocation too aggressive when the pain starts and also why I find optimizing a basket of strategies via some mean variance optimization to be reckless. You seem to be on the right track on tackling this problem. At the end of the day, after the strategy stops working, you want your bank account to have more purchasing power than when you started trading the strategy.
Seeing more flows into bond-ey type stuff, for example we recently went into a fixed income relative val fund. Cherry picking yield extraction ventures seem to be getting more play, as is typical in this stage of a cycle. Don't see that much going into lever up beta to the max type ventures at least not consciously. Of course, not to say this is what you're selling or what your investor is buying. But then not sure what your true source of alpha is and I would think thats what your investor would be most interested in if indeed drawdown is not terminal.