Forums  > Trading  > Use of modelling intraday seasonality in volatility.  
     
Page 2 of 2Goto to page: 1, [2] Prev
Display using:  

rickyvic


Total Posts: 217
Joined: Jul 2013
 
Posted: 2020-06-04 13:03
Ok a few pointers:
1. garch is good for daily mainly, frameworks that do well in intraday vol clustering and seasonality not sure. However there is strong autocorrelation of squared returns, so garch effects do exist, but its not that simple.
2. look at realized volatility measures (minimum 5 min aggregation) and see if it can help
3. seasonality is not like a feature that is going to help as pointed out by EL, the U shaped durations and volume is clear but I would maybe focus on dummy variables that activate around certain spillover events open and close, news, etc. Anyting scheduled can go in the vol forecast or in a dummy that triggers some behaviour or an indicator variable.
4. you do need to forecast volatility, if you are good at it you can have significant changes in your sharpe/sortino/calmar. Having a risk scaling measure is essential to make your pnl volatility as constant as possible.
5. it's not easy I am kind of struggling with it too

A lot of pointers here and you have material to dig further.



"amicus Plato sed magis amica Veritas"

ahgt_123


Total Posts: 9
Joined: May 2020
 
Posted: 2020-06-16 18:22
Just finished reading half of marcos lopez's book advances in financial machine learning.
It is an exceptional book which covered most of my doubts about applying machine learning to stock data.

I have been skeptical of machine learning methods mainly due to advise by seniors and others that they overfit and that if i think logically i myself will discover exploitable patterns.

I am now thinking of starting to apply simple algos like CART or randomForests on intraday data.
Any Inputs ??

rickyvic


Total Posts: 217
Joined: Jul 2013
 
Posted: 2020-06-19 23:20
The more I learn about the machine learning literature the more I think it is a rewrite of what quants have been doing for the last 20 years.

Only exception is deep learning, but I see it difficult to port into our applications, still very interesting... Fascinating almost


"amicus Plato sed magis amica Veritas"
Previous Thread :: Next Thread 
Page 2 of 2Goto to page: 1, [2] Prev