Thank you for the clear explanation. The ice-cream sales over the years increases. How to factor that in the synthetic data you generated? Also, how to make sure, if needed, that the simulated data is always positive. PS. Clipping would make everything
Thanks for the great videos. I am a bit confused here. According to your introduction, it seems that the ARCH(or GARCH?) model is used to model the volatility of the deviance, however it looks like it is used to model the time series itself, rather than the deviance. So my problem is what kind of problem does the ARCH(GARCH) model deals with?
You are doing a great job. Could you suggest me a book that I can learn the basics of model formulation? I am in Turkey. My background is Finance. Your advice would be highly appreciated.
Again my question: How does this apply in reality? Is there some kind of expansion that also makes use of patterns in the sign of the volatility? Lets say I want to predict demand for product X. Say demand in the (t-1) Period was 100, my GARCH model tells me volatility will be high in the next period, I flip a coin for the sign and it lands on positive 1, so I predict 150, and then in reality demand is only 50. Would it make sense to transform the series to something like this: y(t) = abs(mean - x(t)) so considering the deviation of my time series from the mean as a new time series and apply my GARCH model to that? But even then I kinda have the sign problem when using the abs()
Hi Ritvik, thanks again for another piece of great content. I wonder if the reason behind checking the PACF of squared values of simulation formula (not the sqrt value) is to make sure we won't impact the auto correlations at different lags? (because if we don't square both sides of the formula and check PACF directly, PACF doesn't really imply the correct values of PACF anymore?)
Hi, thank you for the video! Can you also make a video about causality in time series data? So given two time-series how can I tell if the first is causing the second one? Is there something besides the Granger test?
Thanks for this great tutorial. One question: Am I right in my understanding that here, you model the returns themselves (a_t) and their volatilities (sigma_t), which is not the same as modeling the errors (epsilon_t) and their volatilities (sigma_t)? I keep getting confused by the usage of epsilon_t as white noise such as here, and at other times as the initial quantity to estimate when interested in error forecasts... When people talk about "the" GARCH model in finance, which quantity are they talking about then? Modeling the returns and their volatilities, or the errors and their volatilities?
Thank you for the great content. I learnt a lot. I was trying few things after watching the videos and got stuck. Q1. I have a AR(3)-GJR-GARCH(2,2,2) model like below. How can I test if the model has any leverage effect with 5% significance level? Which test do I run? best_gjr_garch = arch_model(in_sample_return[ticker],mean='AR',lags =3 ,vol='GARCH',p=2,o=2,q=2,dist='t').fit(update_freq=5) Q2. I also have a AR(3)-GJR-GARCH-M(2, 2, 2) model, how can In test the impact of risk on expected return at 5% significance level? Please help me with the method, I am really stuck.
How did we come with q=2 ? From the original series how do we find out the dependence with previous volatility, because it won't be possible to find that series?
![[FULL EP.16] เซียนพาลุย "เผือก-ฟรอยด์" กรี๊ดลั่น จนปูหนี | เฮ็ดอย่างเซียนหรั่ง | One Playground](http://i.ytimg.com/vi/FsqUvwS3b-Y/mqdefault.jpg)
Wow, the rolling forecast results are quite surprising. Thank you for the quality content :)
Luck to have someone who can make such videos .... Thank you
Thanks you! It would really greate to to see some videos from you about time series predictions using artificial neural networks.
This episode is SO good.
Thanks so much for this quick and clear explaination
Can't thank you enough for making this video
What a great model. Glad i am here! Subscribed 👍🏻
Thank you for the clear explanation. The ice-cream sales over the years increases. How to factor that in the synthetic data you generated? Also, how to make sure, if needed, that the simulated data is always positive. PS. Clipping would make everything
Amazing one! Thanks for sharing.
The time series data was artificially created. Why the model fit did not return near-exact model coefficients?
Hey great vid again! Any chance you could do a video on multivariate GARCH in the future?
Thanks for the great videos. I am a bit confused here. According to your introduction, it seems that the ARCH(or GARCH?) model is used to model the volatility of the deviance, however it looks like it is used to model the time series itself, rather than the deviance. So my problem is what kind of problem does the ARCH(GARCH) model deals with?
Very nice explanation
Thanks and welcome!
You are doing a great job. Could you suggest me a book that I can learn the basics of model formulation? I am in Turkey. My background is Finance. Your advice would be highly appreciated.
Again my question: How does this apply in reality? Is there some kind of expansion that also makes use of patterns in the sign of the volatility?
Lets say I want to predict demand for product X. Say demand in the (t-1) Period was 100, my GARCH model tells me volatility will be high in the next period, I flip a coin for the sign and it lands on positive 1, so I predict 150, and then in reality demand is only 50.
Would it make sense to transform the series to something like this:
y(t) = abs(mean - x(t))
so considering the deviation of my time series from the mean as a new time series and apply my GARCH model to that? But even then I kinda have the sign problem when using the abs()
Hi Ritvik, thanks again for another piece of great content. I wonder if the reason behind checking the PACF of squared values of simulation formula (not the sqrt value) is to make sure we won't impact the auto correlations at different lags? (because if we don't square both sides of the formula and check PACF directly, PACF doesn't really imply the correct values of PACF anymore?)
Hi, thank you for the video! Can you also make a video about causality in time series data? So given two time-series how can I tell if the first is causing the second one? Is there something besides the Granger test?
Wow, it's actually disgusting the correctness of that model 9:15...
Amazing video!
Thanks for this great tutorial. One question: Am I right in my understanding that here, you model the returns themselves (a_t) and their volatilities (sigma_t), which is not the same as modeling the errors (epsilon_t) and their volatilities (sigma_t)? I keep getting confused by the usage of epsilon_t as white noise such as here, and at other times as the initial quantity to estimate when interested in error forecasts... When people talk about "the" GARCH model in finance, which quantity are they talking about then? Modeling the returns and their volatilities, or the errors and their volatilities?
Thank you for the great content. I learnt a lot. I was trying few things after watching the videos and got stuck.
Q1. I have a AR(3)-GJR-GARCH(2,2,2) model like below. How can I test if the model has any leverage effect with 5% significance level? Which test do I run?
best_gjr_garch = arch_model(in_sample_return[ticker],mean='AR',lags =3 ,vol='GARCH',p=2,o=2,q=2,dist='t').fit(update_freq=5)
Q2. I also have a AR(3)-GJR-GARCH-M(2, 2, 2) model, how can In test the impact of risk on expected return at 5% significance level?
Please help me with the method, I am really stuck.
How did you defined 'train' ? .. its not defined within the code!
Just perfect.
I used the exact same code but a different result is produced, one of the model coefficients is not significant, that's strange.
How did we come with q=2 ? From the original series how do we find out the dependence with previous volatility, because it won't be possible to find that series?
Nice video ...thanks
Thank you.
Thank you so much.
Hi, would you mind to point me out on the statistical equation on volatility prediction-rolling forecast? Thanks
Maybe someone can tell me, what Machine learning algorithm is using Python when counting parameters? ANN RNN or some other?
How'd I code a GARCH (1,1) model?
Thanks a lot :-)