When running the worksheet the beta co-efficient produced when fitting GARCH to the simulated time series with the shock value is no longer close to 0 as in the video, instead it's 0.89 as you suggested it should be (and the spike is no longe present in the plot of conditional volatility of simulated returns). Not sure why....
Great video mate, doing my Master's research on measuring the impact of renewable generation on spot market volatility for power markets using GARCH. This is super useful for understanding why it's so commonly used! Do you have a link to any PhD research you did?
Hi mate, glad you are enjoying GARCH as much as I do. Here is my link to my published paper: ses.library.usyd.edu.au/bitstream/handle/2123/14728/2016_Christian_Contino_Thesis.pdf
Where did you do you PhD, and was your research into statistical/mathematical finance? Your videos are very helpful, have you considered adding any advanced material? I liked your video on copulas as well but would love to understand it at a deeper level through your intuitive explnation.
Hi John. I did my PhD at the University of Sydney in Australia, and my focus was intraday data to make better volatility predictions with models such as GARCH. I have a vague roadmap in my mind for advanced material, but want to build some good foundations first to get a good community around my videos. I am also working on building better graphics as well. I am in awe of projects like 3 blue 1 brown, so would like to get to that sort of level, where even complex things become intuitive. Thanks for watching, really appreciate it
Coolest content, amazing vibes. I'm modeling exchange rate volatility with a garch(1.1) using Python ( purpose is pricing options) after reading so many papers and watching a lot of videos I'm actually confused regarding one thing: when using the arch package to estimate a garch model, some include in the function the returns series ( as in this video where we used Apple returns) while others estimate first a mean equation ( let's say using an ARMA model ) and then they take the residuals and use them in the garch function. Is this the same thing? What would be more appropriate?
Hey mate. Thanks for the kind words. So you can use a demeaned model etc, but the difference is negligible. For simplicity I would just go straight to GARCH and be done. You can refine these things later on, but don’t complicate your life. Keep up the good work and look forward to hearing how your project goes. I would usually direct you to my forum but my server is down and will be back next month.
Hi Matthew, GARCH only works at very low forecasts horizons, such as daily or weekly. The process quickly converges to long term volatility Sigma, which is just sample STD. You can’t do multi year forecasts with this.
Great video, thanks for sharing!
Glad you like it mate!
Great job!! Greetings from Brazil
Thank you!
Thank you. Great video. Learnt heaps!
Glad it helped
Excellent. Incredibly you've managed to make econometrics seem hip.
Haha. Too kind sir.
When running the worksheet the beta co-efficient produced when fitting GARCH to the simulated time series with the shock value is no longer close to 0 as in the video, instead it's 0.89 as you suggested it should be (and the spike is no longe present in the plot of conditional volatility of simulated returns). Not sure why....
Love the chill vibes man. Whats that song you got going?
Helped a lot to better understand GARCH, can you please cover volatility measures on panel data as well.
Thanks for watching. Will add it to the to do list
Great stuff,!
Thanks sir!
Great video mate, doing my Master's research on measuring the impact of renewable generation on spot market volatility for power markets using GARCH. This is super useful for understanding why it's so commonly used! Do you have a link to any PhD research you did?
Hi mate, glad you are enjoying GARCH as much as I do. Here is my link to my published paper:
ses.library.usyd.edu.au/bitstream/handle/2123/14728/2016_Christian_Contino_Thesis.pdf
Thanks appreciate it - looking forward to more of your videos
Thank you Sir.
Where did you do you PhD, and was your research into statistical/mathematical finance? Your videos are very helpful, have you considered adding any advanced material? I liked your video on copulas as well but would love to understand it at a deeper level through your intuitive explnation.
Hi John. I did my PhD at the University of Sydney in Australia, and my focus was intraday data to make better volatility predictions with models such as GARCH. I have a vague roadmap in my mind for advanced material, but want to build some good foundations first to get a good community around my videos.
I am also working on building better graphics as well. I am in awe of projects like 3 blue 1 brown, so would like to get to that sort of level, where even complex things become intuitive. Thanks for watching, really appreciate it
Coolest content, amazing vibes.
I'm modeling exchange rate volatility with a garch(1.1) using Python ( purpose is pricing options) after reading so many papers and watching a lot of videos I'm actually confused regarding one thing: when using the arch package to estimate a garch model, some include in the function the returns series ( as in this video where we used Apple returns) while others estimate first a mean equation ( let's say using an ARMA model ) and then they take the residuals and use them in the garch function.
Is this the same thing? What would be more appropriate?
Hey mate. Thanks for the kind words.
So you can use a demeaned model etc, but the difference is negligible.
For simplicity I would just go straight to GARCH and be done. You can refine these things later on, but don’t complicate your life.
Keep up the good work and look forward to hearing how your project goes.
I would usually direct you to my forum but my server is down and will be back next month.
Sure thing. Thanks a lot!
Is the process the same for shorter horizon volatility forecasting versus a longer term forecast for capital market assumption purposes (5-10+ years)?
Hi Matthew, GARCH only works at very low forecasts horizons, such as daily or weekly. The process quickly converges to long term volatility Sigma, which is just sample STD.
You can’t do multi year forecasts with this.
Great videos, would be better if the code was larger. It’s tough to see on mobile
Thanks mate. Yeah, will try and make it super zoom mode next time
This guy is too good looking to have a PhD in statistics
You make me blush