Skip to content

Volatility clusters

8 messages · kafkaz, kafkaz2, Patrick Burns +1 more

Original
kafkaz
# 
Dear all,
what would you propose for volatility clusters identification? 
K-mean doesn't fit because it requires to fix length of clusters. SOM
doesn't look good as well.
I found one paper, where the author refers to F. Laurini works:

F. Laurini, J.A. Tawn (2003), ?New Estimators for the Extremal Index and
Other Cluster
Characteristics,? Extremes, Vol 6, 3, 189-211.

F. Laurini, 2004, ?Clusters of Extreme Observations and Extremal Index
Estimate in
Garch Processes,? Studies in Nonlinear Dynamics & Econometrics, Vol 8, Issue
2, 1-21.

What would be your recommendation?
Patrick Burns
# 
I don't understand the question.

What are you ultimately trying to do?
On 19/07/2010 11:39, kafkaz wrote:

  
    

      
      
    

  


  


      

    

    
1 day later

    

      
      

        


  

        

      kafkaz2
    

    

      
      #
    

  


  

      
The fact is that volatility moves in clusters - high movements are followed
by high movements and low by low. My goal is to identify existing volatility
regime based on historical data. My question is what statistical methods can
I use to map historical volatility data into clusters.

I would expect something like this:

  
    

      
      
    

  


  


      

    

    

      
      

        


  

        

      Patrick Burns
    

    

      
      #
    

  


  

      
Have a look at the finance task view where
you will find packages that do GARCH estimation
and stochastic volatility.

        
On 21/07/2010 07:42, kafkaz2 wrote:

            

      
      
      
    

        

  
    

      
      
    

  


  


      

    

    

      
      

        


  

        

      kafkaz2
    

    

      
      #
    

  


  

      
Dear Patrick,
thank you for the replay.
Actually the graph above was generated by using garch estimation. I have a
broad understanding how garch estimation is done, but I can't understand how
can it identify volatility clusters. The output of this methodology is
predicted variance, but it doesn't say to which volatility regime it
belongs.

  
    

      
      
    

  


  


      

    

    

      
      

        


  

        

      Patrick Burns
    

    

      
      #
    

  


  

      
Regimes are artificial additions -- they have
no objective reality.  That doesn't mean they
can't sometimes be useful.

This gets us back to the question I asked yesterday:
what are you trying to do, and why do you think
that imposing regimes would help you?

        
On 21/07/2010 09:29, kafkaz2 wrote:

            

      
      
      
    

        

  
    

      
      
    

  


  


      

    

    

      
      

        


  

        

      kafkaz2
    

    

      
      #
    

  


  

      
I'm trying to test a claim, that mean reverting strategies work best, then
volatility is high. So far, I improved the results on the strategy by
applying simple volatility filter.
If 5 days volatility is below 2 years 0.35 quantile is stops trading. 
So, my question is - can the filter be more sophisticated?

  
    

      
      
    

  


  


      

    

    

      
      

        


  

        

      Whit Armstrong
    

    

      
      #
    

  


  

      
You should look at the Kim/Nelson book:
http://www.amazon.com/State-Space-Models-Regime-Switching-Gibbs-Sampling/dp/0262112388/ref=sr_1_1?ie=UTF8&s=books&qid=1279726296&sr=8-1

There are also a ton of papers on this topic.

-Whit

        
On Wed, Jul 21, 2010 at 10:36 AM, kafkaz2 <kafka at centras.lt> wrote: