Since there are 10 simulation trials and we are using the highest loss stimulated, wouldn't it be at a confidence level of 90% instead of 95% as mentioned towards the end of the video.
Actually the 90.0% is ambiguous. Under Dowd's approach, the 2nd worst corresponds to the 90% confidence. Because (1-α)*n + 1 = Xth, if we select the 9th worst (out of 10) and ask, what is the confidence implied, then (per Dowd) the answer is 1 - (2-1)/10 = 90.0%. If asked for the 90%, Jorion would give the worst loss. But that's from the direction of imputing confidence, not STARTING with the confidence: if we have discrete n = 10, and ask about the distribution "what is the 95% VaR?" then it is the worst loss because the cumulative 0.95 probability falls squarely "within" the 10th quantile. Under discrete n = 10, the worst loss is the answer to any VaR confidence level greater than 90.0% because the worst loss "occupies" the entire 1.0% cumulative tail. For example, if the question here (n =10) is "what is the 99.9% VaR?", it is also the worst loss (again, from the direction of asking about the quantile GIVEN the confidence level, rather than inferring the confidence from a quantile, which is ambiguous "between the discrete bins" at 0.90)
Thank you for the video. For the next ones, could you please display the Excel bar formula at the top of the screen? It would be easier to follow when you select cells. Thanks!
I wanted to follow-up and let you know that I've incorporated this practice into my videos: for a while now, my videos (when it's possible) include the formula bar. Great feedback!
Thanks for your clear explanations, just; Why we convert prices into returns and again convert them into prices? Why don't we estimate prices directly?
Very interesting and informative video. You state that the file on excel is available, but the link don't have the letter s. It is safe in accessing it? Thank you for teaching
For calculating VaR of the portfolio, is it correct to calculate daily returns of the whole portfolio at the beginning and then resample from those returns? (instead of doing this process for all the stock in the portfolio) since you recommended that we select all the returns in a day together
@@bionicturtle Anytime David!, I did Bionic Turtle curse about 4 - 5 years ago and I loved it, I pass both exams at my first attempt, Every time someone is asking me for advice in taking the FRM exam I always refer them to your program!, keep it up the good work!!!!!
Can you post a link to add in you used to retrieve the stock prices? What's the best way, in your opinion, to include auto correlation to the bootstrap so you can capture correlation between the days? rho * return (t-1) + (1-rho) return(t)?
Hi Anthony, here's is Quandl's excel add-in www.quandl.com/tools/excel Re: improving the bootstrap to capture autocorrelation, it appears that both Kevin Dowd (FRM author) and Carol Alexander (my favorite) recommend Filtered Historical Simulation which combines the bootstrap with (eg) a GARCH volatility model but I haven't coded it (yet) and don't quite understand (GARCH has non-trivial autocorrelation features so I can't say I fully understand how FS GARCH captures autocorrelation). Thanks!
Since formula bar is not visible , i am not able to understand how you pull the portfolio returns against random numbers....Reply is highly anticipated...
The XLS is in the description (www.dropbox.com/s/2iuz39z56auue1j/10-18-bootstrap.xlsx?dl=0) but it's straightforward: =INT(RAND()*21)+1 generates a random integer from 1 to 21 because we have 21 "indexed" actual returns in the historical window. Then VLOOKUP(random index integer ...) retrieves the actual return vector from that day. So INT(RAND()*n)+1 performs the random sampling (with replacement) from the historical window of (n) indexed return vectors.
I guess there is an catch here because you’re simulating 10 trials for the next day. It is not that your simulating for the 10 coming days. If you need to simulate for day 10, then you need to consider T as a multiplier for the variance and therefore the variance will be larger. Therefore the Var might be even worse in day 10.
Since there are 10 simulation trials and we are using the highest loss stimulated, wouldn't it be at a confidence level of 90% instead of 95% as mentioned towards the end of the video.
Actually the 90.0% is ambiguous. Under Dowd's approach, the 2nd worst corresponds to the 90% confidence. Because (1-α)*n + 1 = Xth, if we select the 9th worst (out of 10) and ask, what is the confidence implied, then (per Dowd) the answer is 1 - (2-1)/10 = 90.0%. If asked for the 90%, Jorion would give the worst loss. But that's from the direction of imputing confidence, not STARTING with the confidence: if we have discrete n = 10, and ask about the distribution "what is the 95% VaR?" then it is the worst loss because the cumulative 0.95 probability falls squarely "within" the 10th quantile. Under discrete n = 10, the worst loss is the answer to any VaR confidence level greater than 90.0% because the worst loss "occupies" the entire 1.0% cumulative tail. For example, if the question here (n =10) is "what is the 99.9% VaR?", it is also the worst loss (again, from the direction of asking about the quantile GIVEN the confidence level, rather than inferring the confidence from a quantile, which is ambiguous "between the discrete bins" at 0.90)
Thanks a lot for your prompt response. Its very helpful.
Thank you for the video. For the next ones, could you please display the Excel bar formula at the top of the screen? It would be easier to follow when you select cells. Thanks!
Yes, thank you for helpful feedback, I will try to include the bar formula going forward (if i can). Thank you!
I wanted to follow-up and let you know that I've incorporated this practice into my videos: for a while now, my videos (when it's possible) include the formula bar. Great feedback!
Thanks for your clear explanations, just;
Why we convert prices into returns and again convert them into prices? Why don't we estimate prices directly?
Very interesting and informative video. You state that the file on excel is available, but the link don't have the letter s. It is safe in accessing it? Thank you for teaching
For calculating VaR of the portfolio, is it correct to calculate daily returns of the whole portfolio at the beginning and then resample from those returns? (instead of doing this process for all the stock in the portfolio) since you recommended that we select all the returns in a day together
Very good video Thank you!
Thank you Facundo, we appreciate your support!
@@bionicturtle Anytime David!, I did Bionic Turtle curse about 4 - 5 years ago and I loved it, I pass both exams at my first attempt, Every time someone is asking me for advice in taking the FRM exam I always refer them to your program!, keep it up the good work!!!!!
Can you post a link to add in you used to retrieve the stock prices? What's the best way, in your opinion, to include auto correlation to the bootstrap so you can capture correlation between the days? rho * return (t-1) + (1-rho) return(t)?
Hi Anthony, here's is Quandl's excel add-in www.quandl.com/tools/excel Re: improving the bootstrap to capture autocorrelation, it appears that both Kevin Dowd (FRM author) and Carol Alexander (my favorite) recommend Filtered Historical Simulation which combines the bootstrap with (eg) a GARCH volatility model but I haven't coded it (yet) and don't quite understand (GARCH has non-trivial autocorrelation features so I can't say I fully understand how FS GARCH captures autocorrelation). Thanks!
Thanks so much!
Since formula bar is not visible , i am not able to understand how you pull the portfolio returns against random numbers....Reply is highly anticipated...
The XLS is in the description (www.dropbox.com/s/2iuz39z56auue1j/10-18-bootstrap.xlsx?dl=0) but it's straightforward: =INT(RAND()*21)+1 generates a random integer from 1 to 21 because we have 21 "indexed" actual returns in the historical window. Then VLOOKUP(random index integer ...) retrieves the actual return vector from that day. So INT(RAND()*n)+1 performs the random sampling (with replacement) from the historical window of (n) indexed return vectors.
very nice video isa topic pe mene already detail video bana rakhi h
I guess there is an catch here because you’re simulating 10 trials for the next day. It is not that your simulating for the 10 coming days.
If you need to simulate for day 10, then you need to consider T as a multiplier for the variance and therefore the variance will be larger. Therefore the Var might be even worse in day 10.