Roundup: Six Ways to Create Better Risk-Adjusted Returns
Just a little housekeeping. This is a roundup of my posts over the last week re: creating strategies with better risk-adjusted (smoother) returns.
I’ve also added a sixth to the roundup that I missed previously but might be applicable to some strategies – the abnormal market filter.
Topics covered: (1) Trade low/negatively-correlated assets, (2) trade confidence-based rather than transactional strategies, (3) trade in multiple time frames, and (4) trade multiple conceptually-different strategies.
>> Method #5: Duplicate signal on multiple data sources and indicators
>> Method #6: Abnormal market filter (posts here and here)
I’ve talked a lot over the last few months about my mechanism to determine when the market is acting “abnormally”, the Abnormal Market Filter. The idea behind the filter is to give my own programs a way to measure when the market has moved beyond normal ranges, because strategies based on historical norms might not be suited for that very abnormal moment.
We employ the idea in both our YK and Scotty strategies, but I don’t think the concept is necessarily appropriate for all strategies. In a previous post Random Strategy Development Thoughts I outlined how the filter might work differently (or not work at all) for different types of strategies.
Last comment – just to reiterate what I’ve written previously, none of the six ideas we’ve talked about in these posts is going to make a bad strategy into a good one, just hopefully take a good strategy and make it a little better.
Happy Trading,
ms
To stay up to date with what’s happening at the MarketSci Blog, we recommend subscribing to our RSS Feed or Email Feed.
Filed under: Trading Strategies | 8 Comments
Hi Michael,
Love the blog and the algorithms!
Disclaimer: newbie prattling. This is a simple (and gross) way to reduce risk. And yes, I know it goes directly against the confidence-based v. transactional theories that are so well documented in the Marketsci blog.
The idea: If you have multiple EOD algorithms running (like Scotty, YK, etc), slap an n-day moving average (MA) on each algorithm’s account balance. If the balance slips below the MA, keep tracking the algorithm’s performance, but stop trading it until the account balance rises above your MA. You can either go to cash or (better?) move the funds to one of the still-performing algorithms.
The intellectual basis for such an approach is simple (if completely unproven): algorithms probably trend in the short-term and mean-revert over the longer-term.
Practical benefits:
1. Acts as a circuit breaker for large drawdowns in a “broken” algorithm, circumventing big drawdowns (the biggie)
2. With said circuit breaker and thus greater confidence, perhaps more funds can be invested in these incredibly profitable algorithms than normally would be
3. Easy to calculate and follow
4. Should degrade performance only slightly, if at all
Remember…newbie prattling.
Thanks, WilliamT
RE to WilliamT: very smart comments. I would call this “timing the strategy”. I’ve done a bit with this in the past, but found it ineffective over long time horizons. The main reason I think is that most effective equity trading models at this moment in history are some flavor of mean-reversion – and inherently with mean-reversion, the more ground a strategy loses, the more likely it is to rebound (like a stretching rubber band). I haven’t tested it enough to say that with certainty (and of course non-MR strategies are a completely different story), and this idea is still on my own todo list to explore further.
Again, very smart comments William. Thanks!
michael
I’ve heard William’s idea described as “trading the equity curve”, and have played with it a little. However, the application I’ve seen described has always been just the opposite of what he suggests — you look for a strategy that is down in equity and trade it, on the theory that it will mean revert, as Michael suggested.
Some traders feel that this info can be used to position size. As the equity rises above its average, start reducing your position size, and do the reverse when it’s below.
I don’t have enough experience with this concept yet to be able to endorse it, but I know some folks that I respect who think highly of it.
Carlos, thanks for the memory jog.
The book I got it form is “Smarter Investing in Any Economy: The Definitive Guide to Relative Strength Investing ” by Michael Carr… “Trading the Equity Curve”, page 130. Good one!
Book review: Very nice treatment of lots of flavors of relative strength measures for the newbie. I thought it was a worthwhile read… the equity curve idea was an extra he showed in reducing the (harrowing) drawdowns of some of the RS systems.
Just to be clear… with the incredible performance of the Marketsci algorithms, I look at this technique only as insurance, rather than boosting return. I’m a scaredy cat, and this allows me to invest more $$$ in Marketsci than I would otherwise. So, I guess in a practical sense, it does boost my return.
if an ugly strategy (albeit an unprofitable one) is a hedge, perhaps an insurance policy, then it’s useful if there is some combination of the two that increase reward:risk parameters.
RE to Jeff: I agree, but the strategy would look very different from a traditional trading strategy…something similar to the “disaster overlay” I’ve talked about on this blog or that Taleb HF running around that loses money for years before exploding to the upside during high vol. michael
WilliamT, there are other ways (but not necessarily better) to accomplish what you are speaking of.
One way is to look at win% over N trades compared to historical win%. This can be combined with a ratio of choice such as average win over N trades : average win over all historical trades.
Preliminary testing of this kind of addition to a system where the system is turned off, monitored, and turned back on when it rights itself shows promise.
One way i use to reduce risk is to actually split the trading capital and trade several closely related securitites.
(Normally one reduses risk by trading non-correlated securities.)
If trading an index i try to find three or more tradable securities tracking the same index or an index on the same market. Then i simply trade every security with equal part of the capital.
Every security runs it’s own model.
Every trade generated for every security is taken.
I use the same model for all securites.
Sometimes slightly optimized on each security.