Algorithmic Trading Group Forum
http://www.algotradinggroup.com/cgi-bin/yabb2/YaBB.pl
Trading Strategies >> General Discussions About Strategies >> Regime Switch Modelling
http://www.algotradinggroup.com/cgi-bin/yabb2/YaBB.pl?num=1234788262

Message started by Random on 02/16/09 at 12:44:21

Title: Regime Switch Modelling
Post by Random on 02/16/09 at 12:44:21

I have recently been told by someone at a reputable fund that regime-switching is the holy grail of algorithmic trading.

Has anyone used these before (or looked into it)?

I guess the main problem is lack of enough data.

**** Oops, I posted this before realising that maybe this belongs in the Research section ****

Title: Re: Regime Switch Modelling
Post by Algo Designer on 02/20/09 at 07:00:18

I use it in my high-frequency strategies as a trigger to retrain the adaptive model (partially or solely) or pull out from the market.

How can it be the holy grail though?

Title: Re: Regime Switch Modelling
Post by drin on 02/28/09 at 22:40:37

biocomp has a software for trading system with regime switching strategies http://www.biocompsystems.com/products/finance/profit/index.html
it is at least 10 years old...
Take a look!



Title: Re: Regime Switch Modelling
Post by Random on 03/05/09 at 22:29:04

My guess is they were having problems coming up with reliable signals to indicate regime-switches, but if they could get it to work - they have lots and lots of ideas they would like to try out.

Title: Re: Regime Switch Modelling
Post by statstrader on 03/31/09 at 21:04:30


Algo Designer wrote on 02/20/09 at 07:00:18:
I use it in my high-frequency strategies as a trigger to retrain the adaptive model (partially or solely) or pull out from the market.

How can it be the holy grail though?



Algo Designer,
Care to elaborate on what technique you are using to identify regime changes?  There are quite a few out there, I was just curious to hear yours. Thanks.

Also, could you explicitly state what the types of regimes you are identifying are (mean, parameter, vol,  etc..)?

Title: Re: Regime Switch Modelling
Post by Algo Designer on 04/01/09 at 12:03:40

Hi statstrader,

I use a group of adaptive real-time non-linear predictors that, if they "survive" for long enough, are given an opportunity to feed "quant" signals into more traditional trading bots that take care of the execution strategies. The primary subject of my regime shift monitoring (also real-time) is the statistical significance of those signals. In a sense, I am referring to a very short-term projections of the volatility distribution of the underlying time series, but not exactly. As to the techniques employed, I try to keep things fairly simple by resorting to sequential t-tests tweaked for the problem domain. The monitoring helps determine the extent of the "knowledge" or "experience" deterioration in individual models and timely dispose them in favour of better ones which is a part of the continuous risk management process.

The secondary application is regime shift detection in the mean where I use custom artificial neural network topologies along with a few other statistical shortcuts (a set of white noise filters) to run fast non-linear regressions.

Title: Re: Regime Switch Modelling
Post by Algo Designer on 04/01/09 at 12:32:57


drin wrote on 02/28/09 at 22:40:37:
biocomp has a software for trading system with regime switching strategies http://www.biocompsystems.com/products/finance/profit/index.html
it is at least 10 years old...
Take a look!


Quite a nice facade to a black box. I wonder what they employ... stochastic diffusion, GAs, ANNs, SVMs. What's interesting IMHO is that they manage to do seemingly complex optimisations on single desktop machine! Unless I am completely wrong and one needs to connect that GUI to a grid. I cannot see how one can do decent simulations on one, two, four, eight cores. Modern computers are still too slow! :(

It's strange that they have chosen .NET for their technology, but I might be biased :)

Title: Re: Regime Switch Modelling
Post by statstrader on 04/01/09 at 19:08:03


Algo Designer wrote on 04/01/09 at 12:03:40:
Hi statstrader,

I use a group of adaptive real-time non-linear predictors that, if they "survive" for long enough, are given an opportunity to feed "quant" signals into more traditional trading bots that take care of the execution strategies. The primary subject of my regime shift monitoring (also real-time) is the statistical significance of those signals. In a sense, I am referring to a very short-term projections of the volatility distribution of the underlying time series, but not exactly. As to the techniques employed, I try to keep things fairly simple by resorting to sequential t-tests tweaked for the problem domain. The monitoring helps determine the extent of the "knowledge" or "experience" deterioration in individual models and timely dispose them in favour of better ones which is a part of the continuous risk management process.

The secondary application is regime shift detection in the mean where I use custom artificial neural network topologies along with a few other statistical shortcuts (a set of white noise filters) to run fast non-linear regressions.


Thanks AlgoDesigner.  It would be interesting to see a survey of RS techniques posted.  I was just looking over chow test statistic used in a recent book.

Title: Re: Regime Switch Modelling
Post by val on 04/15/10 at 15:55:29

Hi, who has used biocomp with success ?

Algorithmic Trading Group Forum » Powered by YaBB 2.2.2!
YaBB © 2000-2008. All Rights Reserved.