Open kinyk opened 3 years ago
You have deleted the HTML about ARMA-GARCH-COPULA-VAR ,so I could only bother you here .sry........................Q^Q
Actually, theoretically, you can find the parameters of all types of Copula using MLE (maximum likelihood estimation) or any other methods (such as Inference for Margin-IFM or Canonical Maximum Likelihood- CML), The reason why I use Kendall tau for the last 3 Copula is that when I use the MLE for these 3 in R, it returns errors and I am so lazy to fix bug @-@, so there is no special intention here.
Vào CN, 14 thg 3, 2021 vào lúc 12:57 kinyk @.***> đã viết:
You have deleted the HTML about ARMA-GARCH-COPULA-VAR ,so I could only bother you here .sry........................Q^Q
— You are receiving this because you are subscribed to this thread. Reply to this email directly, view it on GitHub https://github.com/SUSANKI/DADA-MINING/issues/1#issuecomment-798839995, or unsubscribe https://github.com/notifications/unsubscribe-auth/AQIG3KMXPHPSPDCZEDMZI63TDRF6NANCNFSM4ZETYHYQ .
As I and my group members are participating in a research competition, hence I have changed my repository's visibility into private. I will turn it again to "public" after the end of the competition(About 1 month later) for privacy reasons.
Vào CN, 14 thg 3, 2021 vào lúc 16:50 Phụng Nguyễn Ngọc < @.***> đã viết:
Actually, theoretically, you can find the parameters of all types of Copula using MLE (maximum likelihood estimation) or any other methods (such as Inference for Margin-IFM or Canonical Maximum Likelihood- CML), The reason why I use Kendall tau for the last 3 Copula is that when I use the MLE for these 3 in R, it returns errors and I am so lazy to fix bug @-@, so there is no special intention here.
Vào CN, 14 thg 3, 2021 vào lúc 12:57 kinyk @.***> đã viết:
You have deleted the HTML about ARMA-GARCH-COPULA-VAR ,so I could only bother you here .sry........................Q^Q
— You are receiving this because you are subscribed to this thread. Reply to this email directly, view it on GitHub https://github.com/SUSANKI/DADA-MINING/issues/1#issuecomment-798839995, or unsubscribe https://github.com/notifications/unsubscribe-auth/AQIG3KMXPHPSPDCZEDMZI63TDRF6NANCNFSM4ZETYHYQ .
In order to backtest to test whether your VaR calculation is correct or not (ex: whether VaR 95% cover 95% of total number of days), we use several types of tests such as Coverage test, Independence test or Distribution test). I haven't heard of testing the VaR backtesting results using AIC or BIC as normally we don't use them for backtesting (I don't know the reason why). However, I'm sure that the tests I cover in the file (Kupiec and christofersen) are all good and unbiased test to measure the effectiveness of each Copula (as I follow the recommendation from an expert -) For more info in tests in backtesting VaR, u can read these websites for more information: https://www.value-at-risk.net/backtesting-distribution-tests/ http://ceaul.org/wp-content/uploads/2018/10/TR_17_CEAUL_Maio2010.pdf
Vào CN, 14 thg 3, 2021 vào lúc 16:57 Phụng Nguyễn Ngọc < @.***> đã viết:
As I and my group members are participating in a research competition, hence I have changed my repository's visibility into private. I will turn it again to "public" after the end of the competition(About 1 month later) for privacy reasons.
Vào CN, 14 thg 3, 2021 vào lúc 16:50 Phụng Nguyễn Ngọc < @.***> đã viết:
Actually, theoretically, you can find the parameters of all types of Copula using MLE (maximum likelihood estimation) or any other methods (such as Inference for Margin-IFM or Canonical Maximum Likelihood- CML), The reason why I use Kendall tau for the last 3 Copula is that when I use the MLE for these 3 in R, it returns errors and I am so lazy to fix bug @-@, so there is no special intention here.
Vào CN, 14 thg 3, 2021 vào lúc 12:57 kinyk @.***> đã viết:
You have deleted the HTML about ARMA-GARCH-COPULA-VAR ,so I could only bother you here .sry........................Q^Q
— You are receiving this because you are subscribed to this thread. Reply to this email directly, view it on GitHub https://github.com/SUSANKI/DADA-MINING/issues/1#issuecomment-798839995, or unsubscribe https://github.com/notifications/unsubscribe-auth/AQIG3KMXPHPSPDCZEDMZI63TDRF6NANCNFSM4ZETYHYQ .
" if I want to compare the VaR sizes of four stocks, how should I code in this case? "
I can only give u ideas, try it out by yourself.
Choosing the best Copula using backtesting tests (Kupiec,...), let's say it is Gaussian one
Using 10000 simulated log-returns of each stock (CTG, MSN, VIC, VNM) by Gaussian copula, then calculate VaR 99%, 95%, 90% for each stock using quantile 1%, 5%, 10% respectively (just as the same way u calculate VaR for portfolio).
Plot VaR value of 4 companies over period of time on the same graph and compare it. (u can compare VaR 99,95 or 90, it depends on u). Got it?
( Actually I do not fully understand whether u want to compare VaR when treating all 4 assets in the same way or based on the weights of each one in the portfolio. If u also want to consider the weight, then just multiply the simulated log-returns in (2) by their weight before calculating VaR🤔)
Vào CN, 14 thg 3, 2021 vào lúc 17:25 Phụng Nguyễn Ngọc < @.***> đã viết:
In order to backtest to test whether your VaR calculation is correct or not (ex: whether VaR 95% cover 95% of total number of days), we use several types of tests such as Coverage test, Independence test or Distribution test). I haven't heard of testing the VaR backtesting results using AIC or BIC as normally we don't use them for backtesting (I don't know the reason why). However, I'm sure that the tests I cover in the file (Kupiec and christofersen) are all good and unbiased test to measure the effectiveness of each Copula (as I follow the recommendation from an expert -) For more info in tests in backtesting VaR, u can read these websites for more information: https://www.value-at-risk.net/backtesting-distribution-tests/ http://ceaul.org/wp-content/uploads/2018/10/TR_17_CEAUL_Maio2010.pdf
Vào CN, 14 thg 3, 2021 vào lúc 16:57 Phụng Nguyễn Ngọc < @.***> đã viết:
As I and my group members are participating in a research competition, hence I have changed my repository's visibility into private. I will turn it again to "public" after the end of the competition(About 1 month later) for privacy reasons.
Vào CN, 14 thg 3, 2021 vào lúc 16:50 Phụng Nguyễn Ngọc < @.***> đã viết:
Actually, theoretically, you can find the parameters of all types of Copula using MLE (maximum likelihood estimation) or any other methods (such as Inference for Margin-IFM or Canonical Maximum Likelihood- CML), The reason why I use Kendall tau for the last 3 Copula is that when I use the MLE for these 3 in R, it returns errors and I am so lazy to fix bug @-@, so there is no special intention here.
Vào CN, 14 thg 3, 2021 vào lúc 12:57 kinyk @.***> đã viết:
You have deleted the HTML about ARMA-GARCH-COPULA-VAR ,so I could only bother you here .sry........................Q^Q
— You are receiving this because you are subscribed to this thread. Reply to this email directly, view it on GitHub https://github.com/SUSANKI/DADA-MINING/issues/1#issuecomment-798839995, or unsubscribe https://github.com/notifications/unsubscribe-auth/AQIG3KMXPHPSPDCZEDMZI63TDRF6NANCNFSM4ZETYHYQ .
THANKS!!!!!! I will put it into practice!!!!!!
Hi, Mis.susanki, I have received your letter , I have benefited a lot from your article, thank you very much, but I would like to ask you two questions about your ARMA-GARCH-COPULA-VAR model right now QAQ. I wonder if you have time to clear up my doubts. Here are my questions:(1) When checking the fitting degree of the copula model, Maxloglikelyhood was adopted by Gussian copula and T copula, while Kendrall was adopted by the latter three copula. If I want to select the best one from the five copula models and use AIC BIC method to test, how should the code run in this case?(2) In the following calculation of VaR, if I want to compare the VaR sizes of four stocks, how should I code in this case?Your help is urgently needed, thanks again for your reply, looking forward to your reply!!^_^