Acrary is a genius!

[Hombre]

Quote from cipherscribe:


This data needs to be normalised and then correlated before decisions can be made about which multiple system strategy works best.

Adrian

IMHO Just run sim trade programs for each system and handpick the best systems .
Computers can't do everything.

 

[bozwood]

Quote from cipherscribe:

Not sure if this is the best place to ask, but I wanted to know what others use to analyse multiple systems. Much of Acrary's work is based around applying multiple systems that work well together (ie uncorrelated systems).

Backtesting results of many systems over a few years can produce huge amounts of data. This data needs to be normalised and then correlated before decisions can be made about which multiple system strategy works best.

Acrary wrote his own code to gather and produce these results. What do others use?

As an example, take Amibrokers' backtester. It produces a line item for each trade, rather than perday or perweek results which could be correlated directly to another backtest result.

What do people who use multple strategies based on non-correlated systems use to perform this type of testing?

Adrian

get a daily equity change for each system and aggregate that change for each week/month for each system and compare the those time periods to each other.

 

[Indrionas]

acrary will not disclose anything, forget about it. There are just too many variables to know what he was talking about, all we can do is just guess but that's pointless I'm referring specifically to the edge/non-edge trading. His multiple systems weighting stuff is quite basic and can be found in many publicly available papers on portfolio construction out there.

 

[mikesmithv]

I think his revalation of not needing an edge tells a lot about what he is doing. It means that rather than depending on something that happens sometimes (gaps, reaching S/R, etc) he relies on markets simply doing what they aways do. Hence there is no "edge", it's just a matter of taking money out of the market as it makes it's normal moves, for example a system which exploits the continuous sequence of up-thrusts and down-thrusts, over-shooting momentary "fair value" and reversing over and over, short on the down-swings and long on the up-swings. Maybe it's just another way of saying "I'm getting good at this".

 

[New2thegame]

This is inline with what I presume he means as well. Of course I'd rather have him say so than to speculate but I was waiting to see if anyone approached it from my point of view. So, in other words, the combination of trend/chop/countertrend systems spread across different markets and/or varying timeframes can allow the equity curve to smooth while allowing each method to occasionally catch profit from drift, reversion etc. When one method is weak, one of the others may be strong, more than countering the losses. That's my take... A little confirmation would be nice. acrary, where for art thou?

Quote from mikesmithv:

I think his revalation of not needing an edge tells a lot about what he is doing. It means that rather than depending on something that happens sometimes (gaps, reaching S/R, etc) he relies on markets simply doing what they aways do. Hence there is no "edge", it's just a matter of taking money out of the market as it makes it's normal moves, for example a system which exploits the continuous sequence of up-thrusts and down-thrusts, over-shooting momentary "fair value" and reversing over and over, short on the down-swings and long on the up-swings. Maybe it's just another way of saying "I'm getting good at this".

 

[cipherscribe]

Quote from bozwood:

get a daily equity change for each system and aggregate that change for each week/month for each system and compare the those time periods to each other.

Bozwood,

I clever sentence and 2 lines of code, and I have precisely what I've been looking for. Thanks for the direction.

 

[acrary]

As a precursor to building Doron, I had to find a way to characterize market activity. All markets are two dimensional (time and price). Of the two, only time is constant. I built some software to look at price action within a fixed period of time (1 day, 2 days, etc.). I expected to not find recurring themes (since I thought most maket action was random). What I found was a fixed number of behaviors that repeated themselves often. I could run a trendfollowing system in a market, find the poor performing periods and look at the classification system to find what would have worked best during the same time period. In a short period of time I was able to build a small number of systems that did almost as well as edge-based systems. The only thing necessary was to check the markets for changes to the basic themes (a much easier approach than checking for edge deterioration). In the end I felt pretty dumb looking for edges when the classification was so obvious. Anyway maybe it'll save someone else several years of looking in the wrong places.

 

[Indrionas]

Quote from acrary:

As a precursor to building Doron, I had to find a way to characterize market activity. All markets are two dimensional (time and price). Of the two, only time is constant. I built some software to look at price action within a fixed period of time (1 day, 2 days, etc.). I expected to not find recurring themes (since I thought most maket action was random). What I found was a fixed number of behaviors that repeated themselves often. I could run a trendfollowing system in a market, find the poor performing periods and look at the classification system to find what would have worked best during the same time period. In a short period of time I was able to build a small number of systems that did almost as well as edge-based systems. The only thing necessary was to check the markets for changes to the basic themes (a much easier approach than checking for edge deterioration). In the end I felt pretty dumb looking for edges when the classification was so obvious. Anyway maybe it'll save someone else several years of looking in the wrong places.

That's what I've been thinking about for a long time but never dared to check out thoroughly since you were so against trading the market character You wrote: "Once you've traded with an edge, you'll never settle for trading with character of market again."

Why is Doron needed? What does it do?

 

[Jerry030]

Quote from acrary:

As a precursor to building Doron, I had to find a way to characterize market activity. All markets are two dimensional (time and price). Of the two, only time is constant. I built some software to look at price action within a fixed period of time (1 day, 2 days, etc.). I expected to not find recurring themes (since I thought most maket action was random). What I found was a fixed number of behaviors that repeated themselves often. I could run a trendfollowing system in a market, find the poor performing periods and look at the classification system to find what would have worked best during the same time period. In a short period of time I was able to build a small number of systems that did almost as well as edge-based systems. The only thing necessary was to check the markets for changes to the basic themes (a much easier approach than checking for edge deterioration). In the end I felt pretty dumb looking for edges when the classification was so obvious. Anyway maybe it'll save someone else several years of looking in the wrong places.

Could you add some kind of statistic or metric of measurement to this?
Terms like "almost as well" can have many meanings to many people.

In contrast, a statement like "using X method my metrics were (Sharpe ratio, max. average drawdown, Win/Loss ratio in gross dollars) , when I switched to Y method and I saw there was only a 3% decline in W/L ratio, I concluded that method Y was better because the extra complexity didn’t make a significant difference in performance".

Thanks in advance for any information you can give to turn the relative into the specific.

Jerry

 

[joe123]

Quote from acrary:

As a precursor to building Doron, I had to find a way to characterize market activity. All markets are two dimensional (time and price). Of the two, only time is constant. I built some software to look at price action within a fixed period of time (1 day, 2 days, etc.). I expected to not find recurring themes (since I thought most maket action was random). What I found was a fixed number of behaviors that repeated themselves often. I could run a trendfollowing system in a market, find the poor performing periods and look at the classification system to find what would have worked best during the same time period. In a short period of time I was able to build a small number of systems that did almost as well as edge-based systems. The only thing necessary was to check the markets for changes to the basic themes (a much easier approach than checking for edge deterioration). In the end I felt pretty dumb looking for edges when the classification was so obvious. Anyway maybe it'll save someone else several years of looking in the wrong places.

Acrary you do a lot of bullshiting my boy

either speak ENGLISH

or don't speak at all. Are you trying to be the next Jack Hershey, that would be very sad.