SERVING THE QUANTITATIVE FINANCE COMMUNITY

CrashedMint
Posts: 2591
Joined: January 25th, 2008, 9:12 pm

### Duh!

QuoteOriginally posted by: outrunI feel so stupid only realizing this just now...I was modelling the VaR of a portfolio of benchmarks (like 40% equity, 60% fixed income). The benchmarks are GBM and so I thought the portfolio would be *non*- lognornmal,.. but it is, because of the rebalancing!Say the "real" VaR is 10m and your crude model calculates $9.8m while your new and much improved model gets you$9.99m. You improved accuracy by 2%. Now management randomly decides that the holding period should not be 3 days but 5: Your VaR changes by 30%.

Gamal
Posts: 2362
Joined: February 26th, 2004, 8:41 am

### Duh!

Who cares how much your VaR is? It's not a traded asset, it's only for regulatory and accounting purposes, the error of 10% is nothing. You can take any distribution you want.Only one thing is important - stability. If your VaR changes too much in one short time period, you may expect questions from your executives.

CrashedMint
Posts: 2591
Joined: January 25th, 2008, 9:12 pm

### Duh!

QuoteOriginally posted by: GamalWho cares how much your VaR is? It's not a traded asset, it's only for regulatory and accounting purposes, the error of 10% is nothing. You can take any distribution you want.Only one thing is important - stability. If your VaR changes too much in one short time period, you may expect questions from your executives.Correct. So using a really cool model is interesting, but ultimately useless. That said there is some value in VaR.

Cuchulainn
Posts: 62608
Joined: July 16th, 2004, 7:38 am
Location: Amsterdam
Contact:

### Duh!

QuoteOriginally posted by: GamalWho cares how much your VaR is? It's not a traded asset, it's only for regulatory and accounting purposes, the error of 10% is nothing. You can take any distribution you want.Only one thing is important - stability. If your VaR changes too much in one short time period, you may expect questions from your executives.What if the VAR value is constant? Can it be an interval?
Last edited by Cuchulainn on September 16th, 2013, 10:00 pm, edited 1 time in total.
Step over the gap, not into it. Watch the space between platform and train.
http://www.datasimfinancial.com
http://www.datasim.nl

CrashedMint
Posts: 2591
Joined: January 25th, 2008, 9:12 pm

### Duh!

QuoteOriginally posted by: CuchulainnQuoteOriginally posted by: GamalWho cares how much your VaR is? It's not a traded asset, it's only for regulatory and accounting purposes, the error of 10% is nothing. You can take any distribution you want.Only one thing is important - stability. If your VaR changes too much in one short time period, you may expect questions from your executives.What if the VAR value is constant? Can it be an interval?VaR are at least rounded to full thousands so if the portfolio is not too big it's very possible that there is no change. in the reported value.

DevonFangs
Posts: 3004
Joined: November 9th, 2009, 1:49 pm

### Duh!

QuoteOriginally posted by: CrashedMintQuoteOriginally posted by: CuchulainnQuoteOriginally posted by: GamalWho cares how much your VaR is? It's not a traded asset, it's only for regulatory and accounting purposes, the error of 10% is nothing. You can take any distribution you want.Only one thing is important - stability. If your VaR changes too much in one short time period, you may expect questions from your executives.What if the VAR value is constant? Can it be an interval?VaR are at least rounded to full thousands so if the portfolio is not too big it's very possible that there is no change. in the reported value.in some banks they even ignore positions with notional smaller than X, where X can be surprisingly large -- not sure how they can monitor the impact of that

Gamal
Posts: 2362
Joined: February 26th, 2004, 8:41 am

### Duh!

QuoteOriginally posted by: CuchulainnQuoteOriginally posted by: GamalWho cares how much your VaR is? It's not a traded asset, it's only for regulatory and accounting purposes, the error of 10% is nothing. You can take any distribution you want.Only one thing is important - stability. If your VaR changes too much in one short time period, you may expect questions from your executives.What if the VAR value is constant? Can it be an interval?Most executives don't care abour VaR. Some look at plots and when they see an unusual picture, they may start being interested - this is the case of a flat line if it wasn't flat before.And - very very rarely they may see your VaR too big and ask to do something with it. Ask Pat about details

chocolatemoney
Posts: 322
Joined: October 8th, 2008, 6:50 am

### Duh!

QuoteOriginally posted by: GamalWho cares how much your VaR is? It's not a traded asset, it's only for regulatory and accounting purposes, the error of 10% is nothing. You can take any distribution you want.Only one thing is important - stability. If your VaR changes too much in one short time period, you may expect questions from your executives.You probably know already all this, but anyway: make sure that the model you cherry pick and its calibration/estimation procedure are in line with what your regulator expects from you. Actually, in some legal contexts and within some jurisdictions, reporting to the regulator a different VaR than the one used internally may be a problem - in particular if the internal VaR is higher than the one sent over to the regulators in a period of meaningful drawdown. After having learned the lesson, and wasted my time dealing with the bureaucrats, I'd just implement the VaR as the regulator wants it...

exneratunrisk
Posts: 3559
Joined: April 20th, 2004, 12:25 pm

### Duh!

QuoteOriginally posted by: chocolatemoneyQuoteOriginally posted by: GamalWho cares how much your VaR is? It's not a traded asset, it's only for regulatory and accounting purposes, the error of 10% is nothing. You can take any distribution you want.Only one thing is important - stability. If your VaR changes too much in one short time period, you may expect questions from your executives.You probably know already all this, but anyway: make sure that the model you cherry pick and its calibration/estimation procedure are in line with what your regulator expects from you. Actually, in some legal contexts and within some jurisdictions, reporting to the regulator a different VaR than the one used internally may be a problem - in particular if the internal VaR is higher than the one sent over to the regulators in a period of meaningful drawdown. After having learned the lesson, and wasted my time dealing with the bureaucrats, I'd just implement the VaR as the regulator wants it...Agree - even more, regulators want total evidence (who did what and when on each single transaction). So, the integration of valuation and data management (copyright Xenomorph) will become indispensable. But to do risk management that Aaron Brown calls "red-blooded", you might want to be interested in cubes of VaRs (instrument contributions, risk factors, ..). And if your methodology and valuations do this wrong, you might get a wrong picture?It is the old story that comes to my mind from total quality management (in, say, process or manufacturing industries) and the ISO quality certificates (late 80s) - the certifiers wanted evidence, whilst we already started controlling the required quality by production feature control. Forced by aerospace, automotive, .. industries.