FRM: Value at Risk (VaR): Historical simulation for portfolio
HTML-код
- Опубликовано: 13 сен 2024
- This example is a portfolio of three stocks: GOOG, YHOO, and MSFT. Process is: 1. I calculated for each stock the historical series of daily periodic returns (bottom left, below). 2. For each historical day (e.g., Friday 7/18), I calculate the portfolio gain/loss as if I held the current portfolio on that day. This is the essence of the idea: run historical returns through the current portfolio allocation. 3. This produces an historical series (right column, green) of simulated portfolio returns. Now I can treat as with the single-asset; e.g., if I want 95% VaR, then I need = PERCENTILE(range, 5%). For more financial risk videos, please visit our website! www.bionicturtl...
very fantastic, a comprehensive video for historical simulation. i have learned alot from a video made 10 years before
@jokermant in the case, the method is the same because the method merely produces an (empirical) distribution. So, you get the same distribution and instead of a quantile (VaR), you take the average loss in the tail (the losses in excess of the VaR). So, CVaR is just another "look" on the produced distribution.
Another simple way to get VaR, thank you very much.
Thank you for this video it was very useful for me. But i have a question, is the VaR that you calculated for 1 day or 10 days?
What if we need 10 day VaR
@trelemerele i don't know what differences you refer to: 5% sig corresponds to 95% confidence. I think the above is correct, the technical issue concerns measurement of the 5% quantile. The normal distribution does not enter into this: one advantage of HS makes no distributional assumption.
Thank yo Bionic!! obviously the model ignores co-relation between the stocks. Could you please make a similar video involving covariance matrix.
@trelemerele yes, okay I see your point. thanks. I do agree there are different quantile measures: IMO, there are three ways to measure the quantile (e.g., Dowd, Jorion which neither interpolate; or yours). But i'm not aware it is settled; e.g., in your example, Dowd would use 3rd observation without interpolation (function =LARGE(array, 1%*251).
I used percentile() simply to avoid using LARGE()/SMALL() functions and keep it simple. I don't know why yours would necessarily be correct over Dowd
why didn't you Sort the numbers before the calculation of Percentile?
@trelemerele the values are sorted low to high, so I used percentile (array, significance) where 5% sig = 95% confidence; i.e., 5% percentile looks to the "bottom" of the list. There is technical issue at this 5%: there are actually three valid 95%/5% answers but that is a nuance
Why you didn't sorted the Portfolio Daily Profit /losses from worst to best?
@dimdoms23 thanks, I appreciate that. In retrospect (this video is 4 years old), I would have called this bootstrapping rather than HS. Simple HS is just a sort of actual returns.
Can you combine the historical simulation with the variance/covariance approach? E.g. calculating historical (relative) VaR figures for 3 assets as shown in the video but the aggreated VaR is not simply a sum but calculated by using historical correlations between the 3 assets?
@bionicturtledotcom, especialy in historical simulation where returns are not normaly distributed the difference could be large... Anyway thanks for answering.
dear @bionicturtledotcom, I agree with you. Thank you for the dialog.
Dear David,
regarding the percentile function of excel you entered in cell J7, it underestimates the VaR figure, impliying a less than the stated confidence level.
Did you use log returns like in your other video? If yes, log returns are not additive in a portfolio and you can't use the sum function. IMO would be more correct to use simple returns in that case anyway since you convert it to $ values, with log returns you could theoretically have $-200 on a stock even though you only invested $100 with the method you used here?
dear @bionicturtledotcom in my understanding 2.5 obs are 1% out of 250 obs. I've seen this in some papers, but this means nothing as long as model validation i.e backtesting isn't done.I don't know whether Dowd's/Jorion's suggestions are based on empirical backtesting? Especially for derivative portfolios, where small moves in market rates could trigger large P/L moves, this could lead to outliers.Another reason for my approach is that in risk mgmt the conservative approach is always preferable.
How do you get daily returns on a proper portfolio
What if you had a short position for some of your assets?
@trelemerele FWIW, i used this for today's video (see channel if interested)
...thanks for making me re-think it. I agree re: the backtest. I like Dowd's approach because (eg) it gives 3/250 such that the 1% tail contains the two discrete losses (1 and 2). However, your point is interesting: that is less conservative than either jorion or interpolations. Ultimately, i can't convince myself that one is logically superior (given discrete distribution).
Thank you so much :)
@bionicturtledotcom,pls can u help me with any simulation of policy enhancement in the oil and gas industry
very useful to me, thank you.
I had been using MarketXLS for this. It really works for me.
Could anyone help me with calculating the percentile on scientific calci?
Thank you for the video. I've found other Historical simulation calculations for portfolio where a variance-covariance matrix is used. I was wondering why?
That is another method that can be used to calculate the VaR. The variance/covariance approach should be used when the distribution of returns is normal. To stay out of trouble for non-normal return distributions the historical approach is prefered.
Thank you for your helpful answer!
@@thomas9982 Can you combine the historical simulation with the variance/covariance approach? E.g. calculating historical (relative) VaR figures for 3 assets as shown in the video but the aggreated VaR is not simply a sum but calculated by using historical correlations between the 3 assets?
hi, can you tell me from where you got the returns percentage ? is it some random number ? or from the daily charts
+Gayathri N Menon I'm pretty sure it's from the daily charts. You get the prices from yahoo finance and you calculate the return
He insisted on the Word "Daily". The Daily returns :D