Multiple systems at the same time

Because both of you guys seem to be pretty good on this stuff, when adding systems, would you prefer to add a negativly correlated system, or a system with zero correlation?

I could see the arguements for either.
 
Quote from horribilicus:

There's a pretty easy way to create very negatively correlated strategies: ...
More...

Yeah, this post is pretty darn good.

Whereas before I had been trading two highly correlated markets using different criteria for taking profits, I am going to have to give this concept a good hard look.

Best Regards,

JJ
 
I'd do the obvious thing: put together the various possible combinations, try them out, and see how they perform. Why rely on an opinion when you can get experimental results instead?

(An opinion such as "Negatively correlated systems are always preferable to zero correlated systems" is just an excuse not to experiment, I believe.)
 
Quote from horribilicus:


(An opinion such as "Negatively correlated systems are always preferable to zero correlated systems" is just an excuse not to experiment, I believe.)
More...


Touche. I'm working on creating systems to trade/add. I'm sure it also depends on what systems you have etc. I was looking for direction in what kind of system would be good to add on. I guess I will get to work on building one of each to see which works better!!!
 
Quote from minmike:

Touche. I'm working on creating systems to trade/add. I'm sure it also depends on what systems you have etc. I was looking for direction in what kind of system would be good to add on. I guess I will get to work on building one of each to see which works better!!!
More...

What I've studied (theory) and what I've seen with years of screen time (real-time) is that the markets (all) work along the following paradigms (I've spent most of my time - 90%+ - with the financials, so I find this model is greatly applicable to them).

Market Cycles
Phase 1: Price Consolidation before Accumulation
Long and Short strategies will work in this type of market.

Phase 2: Accumulation
Long strategies only will work

Phase 3: Price Consolidation before Distribution
Long and Short strategies will work

Phase 4: Distribution
Short strategies only will work
**
(see diagram)
 

Attachments

The market (Spooz) is currently in Phase 3, while the Longs are no longer working as effectively as they have been for several months, the market has not turned down just yet (see chart).

A model such as you have described here (given sufficient work in development) would ideally be able to take advantage of all 4 phases of market action, experiencing losses only when price action becomes too choppy and range bound, so the entry signals of any trading system would also have to take that into consideration.

Time to go to work.

Best Regards, :)

Jimmy Jam
 

Attachments

Quote from horribilicus:

I'd do the obvious thing: put together the various possible combinations, try them out, and see how they perform. Why rely on an opinion when you can get experimental results instead?

(An opinion such as "Negatively correlated systems are always preferable to zero correlated systems" is just an excuse not to experiment, I believe.)
More...

I like "Things that work and make you happy are preferable to things that fail and make you unhappy." But you can feel free to experiment all you like to test that.
 
Quote from minmike:

Because both of you guys seem to be pretty good on this stuff, when adding systems, would you prefer to add a negativly correlated system, or a system with zero correlation?

I could see the arguements for either.
More...
i would reduce the importance of correlation as a
concept. the reason is that you add an additional
layer of abstraction in the mid of your analysis and
this layer kills some information without adding
any. by this i mean that the whole markowitz
efficiency frontier thing, where all this comes from,
is better used as a financial marketing than a true
investment tool.
consider taking two equity curves. each has its
own stats (return, std, draw downs bla bla). now
you calculate the correlation and after that you
take their stats and combine them by using the
correlations. in principle it sounds fine, but it sucks
since correlations consists of two very distinct parts
for you as a trader. the positive correlation in good
times does not make you worry at all. if the
market is volatile and you make money on each
day with your long break out and your short
breakout strategy, each with a profit target, you
LOVE that positive corr.

actually i can cut this short. i do it as horri, i look
at the strategies and see if their blend has better
stats than the single strategies. you have a kind of
corr-embedment by doing so.
 
Quote from minmike:

Because both of you guys seem to be pretty good on this stuff, when adding systems, would you prefer to add a negativly correlated system, or a system with zero correlation?
More...

Neither; but both, plus also positive correlation. Think about that. :D
 
Quote from horribilicus:

There's a pretty easy way to create very negatively correlated strategies: by design, ensure that whenever strategy A is Long, strategy B can be Short or Out, but it cannot possibly be Long. Ensure that A and B usually have opposite positions, and never have the same position.

The most straightforward embodiment of this idea is: create a filter which indicates whether the very long term trend is up or down. MACD(300,50,3) is one example of such a filter. Strategy A is only allowed to accept new entry signals in the direction of the filter (when filter says "down", strategy A can enter Short but it cannot enter Long). Strategy B is only allowed to accept new entry signals in the opposite direction of the filter (when filter says "down", Strategy B can enter Long but it cannot enter Short).

Voila, strategies A and B cannot both enter long (or both enter short) at the same time. Thus their correlation coefficient is somewhere between -1 and 0. In practice I routinely see correlation coefficients between -0.4 and -0.7.

In practice it is desirable for strategy B, the one that trades opposite to the direction of the very long term trend, to be a relatively short term, "opportunistic" strategy: observe a setup, get in, make some money, and quickly get back out. Exploit short term contra moves while swimming upstream against the very long term trend.

Strategy A is much more forgiving, since it trades in the direction of the very long term trend. You have a lot more choices here.

You can add extra logic if you like, to deepen the negative correlation some more. You can make the entry conditions for Strategy A yet more restrictive, from
Code: 
If (A has an enter long signal) AND
(the filter says up) THEN Buy
to
Code: 
If (A has an enter long signal) AND
(the filter says up) AND
(strategy B is not long) THEN Buy
Naturally you'd add a similar third condition to the entry logic of strategy B. This modification handles situations where the filter switches states before a trade exits. Finally, you can incorporate knowledge of B's position and intentions, into A's exit logic, and vice versa.

A simplistic form of this "manufactured negative correlation" approach is used every day in the managed money arena: let A be a "long only" strategy, and let B be a "short only" strategy. Presto, negative correlation. There are hedge fund managers who trade only from the short side; asset allocaters blend them into a mix with other long-bias managers, and reap the benefits of deep negative correlation.
More...

Haven't you done any backtesting yet? Any briefing of performance summary/ data to share with us? :confused:
 
You must log in or register to reply here.
Back
Top