Acrary is a genius!

Status
Not open for further replies.
Variance reduction does not require negative correlation (although that can help).

As long as correlation is less than 1 then you will get variance reduction in the portfolio of strategies/assets/whatever.

Do a search for modern portfolio theory. Better yet play about with it in Excel, use 2 systems to simplify.

Of course, correlation is a spurious measure to start with ...
 
I think I am going to have to side with maestro on this one (hopefully I am interpreting him correctly). If you take 2 structurally independent systems (defined here as different mean/variance characteristics in w/l profiles) and run a monte carlo over a large run of trials of the two systems, you can get a range of the correlations between the two systems, in order to observe how the total risk/reward of the combined system fared over that range of correlations.

Note that the combined system avg reward/risk is simply the average of the two independent system reward/risk averages.

The result of a small run of 200 total trials is shown. Note that the combined reward/risk ratio stays pretty much flat
about the expected avg rwd/rsk of the combined systems (the variance about the average rwd/rsk is due to the joint variance of the independent system variances not due to a systematic improvement).

There is no noticeable systematic improvement in the total avg. w/l of the combined systems over different correlations. You could increase the number of trials to get a larger range of correlations (-1 to +1), but I do not expect the results to deviate much from the conclusion demonstrated.

28lxdhw.jpg
 
Not to get off topic but I never understood how Acrary said he trades all 'strategic' and not tactical.

He went on to explain there is a finite # of behaviors present in the market and then he tried to estimate what behavior was present that day in the market and plug in the specific strategy designed for that behavior.

I see infinite types of behaviors in the market. By that I mean if you were to plot 500 daily sp charts over each other you would end up with a block of ink.

Tactical is the only way I see trading :confused:
 
Quote from dtrader98:
I think I am going to have to side with maestro on this one (hopefully I am interpreting him correctly). ...Note that the combined system avg reward/risk
More...
great analysis there dt, unfortunately, maestro's whole argument is MOOT...

and the reason is: THE OP REFERENCED ACRARY'S WORK WITH NON_CORRELATED SYSTEMS, WHICH HAD NOTHING TO DO WITH R:R, BUT RATHER SHARPE.. IE VOLATILITY OF RETURNS!!!

so NOTHING is in contention here, except for the FANTASY that someone is arguing the other side of the R:R "arguments".

the "argument" stemmed from the fact that man was confused (w/ maestro) because he actually understands acrary's work, while maestro was confused (w/ the entire topic) because he does not. hopefully the both of you (dt and maestro) will do a little reading up on that work, so that we can cut-out the off-topic banter on R:R. you're both obviously smart guys, so maybe stop using those smarts to assume you know what someone else's work is and use them to actually understand it.
 
dtrader

i like when people quickly check stuff and come back.
well done. but from my perspective the flaw starts
earlier on, namely at the utility function. expectation,
or what you call risk/reward, is simply the absolute
wrong thing to look at. it does not tell you anything
about the distribution of outcomes, since it just uses
mean of wins and the mean of losses. the point is you
can have two portfolios with the same value in this figure,
but one would certainly make you a billionaire and the
other one would certainly make you broke.

think of counter trend scalping. tiny expectation (i
hesitate to call it risk/reward, since that name is truly
misguiding), but an equity curve as stable as a rock.
take the same expectation at a trend follower on daily
data and you get bust. huge difference. no way the
figure tells. that means: that figure is the wrong starting
point.
 
it is the absolute same thing with profit factor. completely
useless on a standalone basis. these figures come out
of trader thinking. absolute dollars. absolute amount
won to absolute amount lost. but they need several other
to derive a useful picture. on their own these figures are
meaningless at best, misguiding at worst.
 
Quote from lorndavies:

great analysis there dt, unfortunately, maestro's whole argument is MOOT...

and the reason is: THE OP REFERENCED ACRARY'S WORK WITH NON_CORRELATED SYSTEMS, WHICH HAD NOTHING TO DO WITH R:R, BUT RATHER SHARPE.. IE VOLATILITY OF RETURNS!!!

so NOTHING is in contention here, except for the FANTASY that someone is arguing the other side of the R:R "arguments".

the "argument" stemmed from the fact that man was confused (w/ maestro) because he actually understands acrary's work, while maestro was confused (w/ the entire topic) because he does not. hopefully the both of you (dt and maestro) will do a little reading up on that work, so that we can cut-out the off-topic banter on R:R. you're both obviously smart guys, so maybe stop using those smarts to assume you know what someone else's work is and use them to actually understand it.
More...

ok. Not knocking acrary's work, as I already mentioned I'd like to see more.
Just wanted to try to put the rr argument to rest (and from my perspective there was plenty of contention over the basic concept maestro was arguing, which appeared crystal clear to me). Granted it's slightly off topic, as acrary had some subtle differences in the definitions (structural correlation may be different under his def, and rather than rr, he specifically mentioned re/max dd metric.

Anyways, I'll be happy to refrain if this side menu pontificating is irrelevant.
Look forward to hearing more intelligent discussion on acrary's ideas (particularly from the man himself), whereby maestro's ideas are always welcome.
 
Quote from minmike:

I think you got most of what I was saying. i talk much better than I write.

I have only found non correlated systems. I have yet to find two profitable negatively correlated systems. Have you had better luck with negatively correlated systems than with non correlated systems? I have always looked for non correlated systems (partially because they seem easier to come up with.) Maybe I should start looking for negativle correlated.
More...
the most important thing in system development these days is the
use of human intelliegence in connection with a computer. by that
i mean the early chocie of "what to seek" is utterly important. let
me be more specific. if you run a random system on daily data you
will find about 3% of such runs having a sharpe above 1. i do not
know the figure from top of my head, but you will find some above
2 and some above 3. since today's computer power (i for example
have 56processors combined for backtesting) enables you to run
millions of tests you constantly find tons of flukes. big number of
purely random systems with a sharpe above 3.

so the problem is to first find a way how to distinguish "true" systems
from "fluke". there are several ways we try to tackle this, one of
them being the choice of what we look for. we want to understand
the effect in the first place, using our human intelligence as a prefilter
before we start a single code line.

what i am saying is that the path you choose before you start testing
is utterly important. alan chose statistical analysis of the whole
time series, like the sp future, to derive trading ideas he would
then follow.

if your choice is to find negatively correlated stuff it is first of all
important to completely understand the current stuff. when do
you profit, but more importantly, when do you fail currently. could
be low vola environment. then you need something that makes
money in exactly that siutation. and you go deeper and deeper
into that. what makes money in low vola, but does not get too
heavily caught in high vola. and so forth. you see the point. this
is probably the only way to get into negative corr systematically.
we found that our intraday trading negatively corrs our intraday
stuff. but that negative corr was not intended. just happened.

nevertheless, it is probably smarter to add two uncorrelated systems
than one with negativ corr. the development time for the two
might actually be shorter ...
 
I'm lightly following this thread and I'm waiting for someone else to provide a substantial answer towards this topic but no one seems to so...

I'll BITE.

Dealing with correlated and non-correlated system is a portolio level issue. Theoretically speaking, having a non-correlated system IS far more advantage-ous, in terms of smoothing your equity curve.

Though in real life trading, this isn't the case. You have margin/capital limitations and you need to have a sound portfolio level allocation towards which systems and markets you trade.

Let's say you have two systems:

Intraday scalping system where you are trading 5-10 times a day with a 2.5ish Sharpe and a R:R of 0.75.

Long-"er" term system where you are trading 1-5 times a month with a 0.75 Sharpe and a R:R of 2.5.

Now... both systems have a correlation of 0.33 on a monthly basis.

Under a static size portfolio allocation, the intraday system will be holding a larger weight on the portfolio.

Real trading does not work that way. Portfolio allocation is a lot more dynamic than what is discussed over this thread. Markets are always changing, and "generally speaking" shorter term models are more vulnerable towards change. You can't place most of your money on the short-term model considering how the future may change.

You need to find the perfect balance between maximizing your performance, while REDUCING YOUR RISK.

I'll declare one thing...

Most of you who's posted in this thread are sucker system traders. You guys are too concentrated on Maximizing a certain aspect that you are blinded about thinking in terms of Minimizing the risk. To give a hint on what I mean... R:R is vulnerable towards the freqency of the reward. All measures have a flaw/vulnerability and always have a required counter-measureS that needs to sustain a certain level.

Human aspect is a completely seperate issue towards what's discussed in this thread.
 
Status
Not open for further replies.
Back
Top