All New Wilmott Jobs Board                     (b)

Forum Navigation:

magazine

FORUMS > Trading Forum < refresh >
Topic Title: Volatility using high frequency data
Created On Tue Jun 03, 08 08:24 AM
Topic View:

View thread in raw text format


JediWarrior
Member

Posts: 23
Joined: May 2007

Tue Jun 03, 08 08:24 AM
User is offline

I would like to know if any market practitioner finds interesting the calculation of volatility using tick-by-tick data. In my experience I find that taking into account estimators that take include extreme values, such as Parkinson´s, Garman-Klass, ... is interesting to estimate real volatility. But I have no idea if using tick-by-tick data adds up something. By the way, with these high frequancy data you need new estimators, as they are highly biased.

Any ideas on this? Any recommended papers?

Thanks in advance
 
Reply
   
Quote
   
Top
   
Bottom
     



Paul
Senior Member

Posts: 5419
Joined: Jul 2001

Tue Jun 03, 08 09:07 AM
User is offline View users profile

It depends what you need the volatility for. If it's for trading at high frequency then yes, it is useful. If you need vol for option pricing and you are going to be hedging at a lower frequency, daily, say, then it is less useful. The question then becomes does the vol measured using high-frequency data 'scale' so as to give you a good measure of volatility over longer periods. This is your point about biases.

This is related to the point about using extremes. If a vol measure relies on normality of distributions then using extremes will be a problem. But then it also comes down to your hedging strategy. If you hedge when the underlying moves a lot then data in the intraday extremes is important. If you hedge just before the close every day then daily data, and not extremes or high-frequency data, is important.

P
 
Reply
   
Quote
   
Top
   
Bottom
     



JediWarrior
Member

Posts: 23
Joined: May 2007

Tue Jun 03, 08 09:45 AM
User is offline

1) My goal is intradaily gamma hedging: several times per day, specially if the market moves. Not really high frequancy, but closer.
2) The other goal is estimating the "real" volatility compared to the implied one to enter such strategies such as buying a stranddle ATM to bet for the drift. Or to price a short term product.
3) Using volatility measures that incluse High, Low, such as Garman-Klass helps (the historical volatility obtained is closer to the implied one). I do not know if it is worthy for me to enter the tick-by-tick arena.
 
Reply
   
Quote
   
Top
   
Bottom
     



msperlin
Senior Member

Posts: 608
Joined: Jul 2006

Tue Jun 03, 08 09:46 AM
User is offline View users profile

Quote

Originally posted by: JediWarrior
I would like to know if any market practitioner finds interesting the calculation of volatility using tick-by-tick data. In my experience I find that taking into account estimators that take include extreme values, such as Parkinson´s, Garman-Klass, ... is interesting to estimate real volatility. But I have no idea if using tick-by-tick data adds up something. By the way, with these high frequancy data you need new estimators, as they are highly biased.

Any ideas on this? Any recommended papers?

Thanks in advance


I've been working with ACD models and one of the applications is to compute high frequency volatility. The idea is still quite simple, based on a a la garch type of formulation (lag dependence and clustering), the new model just changes the unit if time for the volatility (and mean equation) at each time t in order to input irregularly spaced data into the model.

Here some references for you to check out:

ENGLE, R, RUSSEL, J. R. (1998) ‘Autoregressive Conditional Duration: A New Model for Irregularly Spaced Transaction Data’ Econometrica, Vol. 66, N. 5, pp 1127-1162.
ENGLE, R. (2000) ‘The Econometrics of Ultra-High-Frequency Data’ Econometrica, Vol. 68, N. 1, p. 1-22.
GERHARD, F. HAUTSCH, N. (2002) ‘Volatility Estimation on the Basis of Price Intensities’ Journal of Empirical Finance, v. 9, p. 57-89.

And a comparison against realized vol is given here:

COEN, A., RACICOT, F. (2004) ‘Integrated Volatility and UHF-Garch Models: A Comparison Using High Frequency Financial Data’ Working Paper, available at: http://ssrn.com/abstract=498222

It seems that realized is better.

-------------------------
My personal site with Matlab Code and research papers here

Edited: Tue Jun 03, 08 at 09:51 AM by msperlin
 
Reply
   
Quote
   
Top
   
Bottom
     



JediWarrior
Member

Posts: 23
Joined: May 2007

Tue Jun 03, 08 10:01 AM
User is offline

Msperlin, do you think that those techniques would improve my life as gamma scalper (and P&L) trading around 10 times per day?
 
Reply
   
Quote
   
Top
   
Bottom
     



msperlin
Senior Member

Posts: 608
Joined: Jul 2006

Tue Jun 03, 08 10:12 AM
User is offline View users profile

Quote

Originally posted by: JediWarrior
Msperlin, do you think that those techniques would improve my life as gamma scalper (and P&L) trading around 10 times per day?


Sorry, I have no idea what you're talking about. I hope scalping doesn't involve killing anyone, which, in this case, ACD models won't be much of a help.

I know what gamma means, but what about scalping it ?? Can you clarify ?

-------------------------
My personal site with Matlab Code and research papers here
 
Reply
   
Quote
   
Top
   
Bottom
     



JediWarrior
Member

Posts: 23
Joined: May 2007

Tue Jun 03, 08 10:27 AM
User is offline

Sorry, I try to explain. Imagine you are long gamma optionality. You can play with the realized volatility against the implied one buying/selling the underlying.
 
Reply
   
Quote
   
Top
   
Bottom
     



msperlin
Senior Member

Posts: 608
Joined: Jul 2006

Tue Jun 03, 08 10:37 AM
User is offline View users profile

Quote

Originally posted by: JediWarrior
Sorry, I try to explain. Imagine you are long gamma optionality. You can play with the realized volatility against the implied one buying/selling the underlying.


I didn`t really quite got it, but thats fine.
If what you need is a measure of volatility for each arrival of a new quote (or trade), then ACD-Garch can give that to you.

Regarding performance in the trading strategy, that has to be properly tested.

-------------------------
My personal site with Matlab Code and research papers here

Edited: Tue Jun 03, 08 at 10:38 AM by msperlin
 
Reply
   
Quote
   
Top
   
Bottom
     



Gmike2000
Senior Member

Posts: 797
Joined: Sep 2003

Wed Jun 04, 08 08:13 PM
User is offline

what underlying is so volatile that it pays to scalp its gamma ten times a day, after paying bid / offer?

not sure if intraday data is helpful for this particular purpose. there are lots of problems associated with it...it can have bad quotes, but there is no good automated way to clean them out (because a jump could be real...). then you have the problem of bid/offer bounce at small time steps. then there is the question of how to treat overnight movements. then there is the question on how to scale your tick volatility into daily vol...is it scalable after all? what if the tick data is not evenly spaced? also, excel ain't gonna do it (the new excel 2007 accomodates 2million or so lines, but still...), even matlab runs out of memory when loading too big of a dataset. it is very tricky to work with intraday data sets and it will certainly distract you from trading during the day.
 
Reply
   
Quote
   
Top
   
Bottom
     

View thread in raw text format
FORUMS > Trading Forum < refresh >

Forum Navigation:

© All material, including contents and design, copyright Wilmott Electronic Media Limited - FuseTalk 4.01 © 1999-2014 FuseTalk Inc. Terms & Conditions