VaR

Discussions about Money Management and Risk Control.
wonkabar
Full Member
Full Member
Posts: 20
Joined: Wed Sep 01, 2004 5:27 pm

VaR

Post by wonkabar » Tue Sep 28, 2004 12:15 pm

I have never looked at this but someone recently asked me what the VaR would be for a diversified LTTF system. Obviously it depends on the markets traded and the risk taken but if someone is trading, say, 20 diverse markets and averages 15% margin:equity is there any way to approximate a VaR number?

Does anyone know of a free web page where a portfolio could be plugged in to calculate this?

kianti
Roundtable Knight
Roundtable Knight
Posts: 335
Joined: Fri May 02, 2003 6:10 am
Location: Florence - Italy

Post by kianti » Tue Sep 28, 2004 12:22 pm

http://www.gummy-stuff.org/VaR.htm
Excel spreadsheet included.

Jake Carriker
Site Admin
Site Admin
Posts: 1493
Joined: Fri Sep 12, 2003 10:32 am
Location: Austin, Texas

Post by Jake Carriker » Tue Sep 28, 2004 4:16 pm

Speaking of VAR, there is an interesting series of articles that debate its usefullness near the bottom of Nassim Taleb's web page at http://www.fooledbyrandomness.com. The links to the articles are under the heading "The Value-at-Risk Debate".

I tend to agree with Taleb's views on VAR, and I have never used it in my own trading.

Best,
Jake

ksberg
Roundtable Knight
Roundtable Knight
Posts: 208
Joined: Fri Jan 23, 2004 1:39 am
Location: San Diego

Post by ksberg » Tue Sep 28, 2004 4:49 pm

If you read and follow the VaR links provided above, it includes links to both PRO and CON articles about using VaR, including interviews with Taleb. In particular, the PRO piece by Philippe Jorion presents a very pragmatic view.

Cheers,

Kevin

wonkabar
Full Member
Full Member
Posts: 20
Joined: Wed Sep 01, 2004 5:27 pm

Post by wonkabar » Tue Sep 28, 2004 6:24 pm

My interest in VaR is along the lines of being curious to what the number would be. I know banks and funds tend to use things like this and think it would be interesting to see what the numbers say. It is just a tool like anything else and people who over rely on the number generated and lose focus as to how and why their portfolios are being constructed will inevitably be in for a shock at some point. The only sure thing is that we do not know what our true risk is and there is no way of measuring it since we can not predict the future.

Jake Carriker
Site Admin
Site Admin
Posts: 1493
Joined: Fri Sep 12, 2003 10:32 am
Location: Austin, Texas

Post by Jake Carriker » Tue Sep 28, 2004 8:27 pm

Ah, I did not see the links I was trying to point to were already included at the site Kevin linked to. Good catch. BTW, the Taleb site also has both the pro and con articles.

Jake

ksberg
Roundtable Knight
Roundtable Knight
Posts: 208
Joined: Fri Jan 23, 2004 1:39 am
Location: San Diego

Post by ksberg » Wed Sep 29, 2004 12:24 am

Jake Carriker wrote:BTW, the Taleb site also has both the pro and con articles.
LOL ... yeah, and I noted that after posting my response. Oh well.

Both sides seem to have reasonable positions. Essentially: (CON) VaR is incorrect and misleading, (PRO) something is better than nothing.

Cheers,

Kevin

sagev
Senior Member
Senior Member
Posts: 31
Joined: Thu Mar 25, 2004 12:49 pm

Post by sagev » Wed Sep 29, 2004 3:59 am

IMHO, I think VaR is not applicable to the LTTF systems in which markets diversification plays an important role and the size of money allocated to every single market has to be allowed to be fluky. Thus even if a constant value of VaR is measurable, it will not significantly improve the performance of a LTTF system. At least in my testing experience I do not find it any good.
Just my 2 cents.

Dan G

Post by Dan G » Sat Nov 20, 2004 4:51 pm

VAR is a decent measure of risk. It does have massive problems, like not accounting for fat tails in return distrubutions, and no standard way of calculating the measure, and the inability to sum VAR across portfolios.

However, it has some uses. Every risk manager will look at this value, so if you are trading OPM, you need to know and use it. Also, knowing what a big loss will be (not a max loss, but something to expect 1 in 20 times) is pretty useful.

Note that VAR for a LTTF system is going to vary depending on how many positions are being held at that time, and the calculation method used (historical, variance/covariance, monte carlo).

Dan G

Post by Dan G » Tue Dec 07, 2004 10:26 am

I've been thinking the only way to get a reasonable number for VAR in a TF system would be to run a monte carlo simulation with proper assumptions on a fully loaded portfolio and calc the VAR. Once this number is in hand, then adjust VAR for the current level of 'loadedness' of the portfolio.

This is the only way I can see that accounts for the dramatic differences in levereage employed by this strategy. Like to hear some other suggestions, if you're thinking on this.

To be honest, I can't do this. Maybe someone else ( a MC expert) can create the strongly and loosely correlated sectors, run the numbers and come up with a percentage VAR. I wouldn't expect them to share this model - it would be pretty valuable to a FOF or family office.

Ragnar D.
Full Member
Full Member
Posts: 11
Joined: Thu Oct 14, 2004 1:58 am

Post by Ragnar D. » Tue Dec 07, 2004 12:19 pm

Any type of measurement that assumes a normal distribution ie: VaR, Sharpe, correlation is faulty if its application is in data that do not come even close to exhibiting a normal distribution.


Claiming to extract profits from the market from its non-random behavior then simultaneously use Gaussian-based metrics poses a contradiction.


Qualitatively understanding the dependency of each instrument, strategy or in Mike's case FOF is vital if one wants highly desirable returns. Unfortunately, the qualitative part cannot be quantified.


ksberg says:

(PRO) something is better than nothing
--No information is better than wrong information: zero information works just fine for me.

Kevin, would you backtest your system with this "something is better than nothing" data? Hope not (':D')




Mike S. says:

I wouldn't expect them to share this model - it would be pretty valuable to a FOF or family office.
--True indeed, it always fades me when a FOF manager dissects a track record and run their "High-Tech Artificial Intelligence Analytics" (especially correlation)...one not too industrious FOF manager I encountered claims the return of a strategy is too volatile albeit the strategy having a MAR of 2. Well, that's his job as a FOF manager... which is to assemble varying asset classes which carries no dependency and create a portfolio that exhibits a smooth return...just like how fund managers do it albeit on a micro level working directly with the individual instruments.


This particular FOF manager expects to get a 20% performance fee having everything delivered to him on a silver platter. ? These FOF people are dangerous.


:)

Dan G

Post by Dan G » Tue Dec 07, 2004 8:29 pm

Ragnar D.-

You could use a non-normal distribution of returns in MC testing, even one based solely on historical returns.

There are many faults with VAR, but I do think its better than nothing. I think the main issue lots of people have with it is fact they want it to be a stress test style measure, and VAR is not a stress test. I am reading a book right now by a justifiably famous HF person, and this person blurrs the distinction several times in the book.

I think that people don't like to understand that 1 in 20 happens all the time, so they make it a stand in for a stress test/worst case loss. For example, this very smart person points to the loss for a particular strategy during the Russian crisis as exceeding VAR. Doh!!! Well a 95% monthly VAR is going to happen 1 in 20 times - thats less than every 2 years if it just was in order! Using some baysean(sic), you can easily determine that there is a great chance of experiencing this 95% VAR in the next year.

VAR just what you could expect to lose some defined percentage of the time. Its not a worst case scenario, or a likely loss under specific bad circumstances. It has lots of assumptions. It has lots of flaws. You pointed several out in your post. But, if you make wise assumptions and understand what VAR means, and its limitations, it can be a decent number to look at and use as a measure of risk in conjunction with other ways of evaluating risk.

Do I think its useful for LTTF? No. Do I think its useful for a FOF? Yep. So if you want an allocation, its good to have a number, and a good methodology that you can explain to this person.

Risk for me isn't one number, its many numbers and experience and common sense and structure. VAR is one small way to get a more complete picture of risk.

Perfection is not attainable, but excellence is.

TC
Roundtable Knight
Roundtable Knight
Posts: 101
Joined: Thu Jul 08, 2004 2:12 pm
Location: Toronto, Canada

Post by TC » Tue Dec 07, 2004 8:59 pm

Mike S wrote:......Risk for me isn't one number, its many numbers and experience and common sense and structure. VAR is one small way to get a more complete picture of risk. ......
Mike

If you care to share, what risk metrics do you find most useful and how do you use/interpret them in trading for your own account ?

Thanks

Tom

shakyamuni
Roundtable Knight
Roundtable Knight
Posts: 113
Joined: Wed Jun 04, 2003 9:44 am
Location: Somewhere, Hyperspace

Post by shakyamuni » Wed Dec 08, 2004 7:48 am

VAR
Last edited by shakyamuni on Sat Jan 08, 2005 5:02 pm, edited 1 time in total.

Dan G

Post by Dan G » Thu Dec 09, 2004 9:05 am

I use all the typical ones - look to the 'How do I know if a system is good thread' for a better discussion. MAR, sharpe, Sortino...

I also use something I cannot find the formula for right now. Actually two formulas. They are more gambling formulas, but come from bayesian statistics. One tells you the odds of having a losing streak lasting x number of times, given a winning percentage. The other tells you the percent odds of losing x times in over the next n samples.

These are useful for me bacause they can show you how your system is performing compared to its average performance. It helps me to keep my head when the system is performing well, and even better, gives me hope when its in a drawdown. By using them frequently with different scenarios or over a range of values, they give you a great feel for what to expect in the future.

When you know there is a 30% chance over the next 100 trades that your win % will be 1/2 its normal level, it tends to make you want to risk less and be more proactive in managing your risk. In addition, during good times, it they help to remind you this is essentially a random and expected good time - its not because the gods are smiling upon my ugly mug.

I also think in market time, not actual time. I take the average number of trades per month, and take this as a month of trading. So when I get that number of trades, its been a 'month', even if these trades happened over just a week. Of course, I also use real time, but I find that thinking in terms of number of trades helps to give me a standardized base of comparision.

In addition, I look to back tested results for periods of under and over performance and look to the win/loss percentage before trading a system.

I guess the reason I focus on this is because I've had those long streaks of losers and winners, and I like to put them into perspective. By looking at what might happen, I think its easier to trade through these times.

Also, it helps to tell if the system is 'broken'


I am not trading right now for a few different reasons, but this is what I looked at when I was actively trading. Thanks for the question. When I find these formulas again, I will post them.

Ragnar D.
Full Member
Full Member
Posts: 11
Joined: Thu Oct 14, 2004 1:58 am

Post by Ragnar D. » Thu Dec 09, 2004 10:04 am

Mike S wrote: I think that people don't like to understand that 1 in 20 happens all the time, so they make it a stand in for a stress test/worst case loss. For example, this very smart person points to the loss for a particular strategy during the Russian crisis as exceeding VAR. Doh!!!
Yes!!!. exactly...most participants are unwilling to feel the probability of losing money...instead they cover it up with smart numbers

...From what i gather from your post, these smart ones also blamed the event to human factors??? They are claiming that 99% of the time he can predict the probability of the distribution of leaves when they fall. But if a human sweeps those leaves into a pile, that doesn't count because that's not part of the normal distribution...What is this guy thinking??? So this person blames his losses on the Russians, not his lack of perception that a risk of default is a high probability.

I dont need to know what the percentage numbers are, all i need to make is a binary decision that it may happen and i wont lose my shirt when it does. Its that simple.


The fortunate thing is most participants will continue to keep using random walk based metrics.

JimR
Full Member
Full Member
Posts: 21
Joined: Sun Feb 27, 2005 5:03 pm
Location: New Jersey

Post by JimR » Sat Mar 12, 2005 4:54 pm

It's been a few months since anyone posted here, but I'll add this anyway. I just read the Taleb interview at derivativestrategy.com and the VAR article on his own site and it seems clear now, at least in my mind, that the entire VAR model & related assumptions are highly flawed.

This exchange between Taleb(NT) & the interviewer (DT) really does sum it up:

DS: Proponents of VAR will argue that it has its shortcomings but it's better than what you had before.

NT: That's completely wrong. It's not better than what you had because you are relying on something with false confidence and running larger positions than you would have otherwise. You're worse off relying on misleading information than on not having any information at all. If you give a pilot an altimeter that is sometimes defective he will crash the plane. Give him nothing and he will look out the window. Technology is only safe if it is flawless.


Think about that... if there is even the slightest chance that you can die, (or in a trader's case, blowup. ie Niederhoffer) using such a model, isn't that particular risk model effectively useless?

The Turtle risk management system, or similar variations thereof, while not perfect (which doesn't exist anyway), will never cause you instant death. :shock:


Any remarks would be welcome.

JG
Contributing Member
Contributing Member
Posts: 6
Joined: Fri Jan 07, 2005 5:48 pm

Post by JG » Sun Mar 13, 2005 8:59 am

CME's SPAN margins and VaR are very similar in concept. The 95% daily VaR for a TF portfolio, as a percent of trading level, in my experience, is usually significantly less than a SPAN'ed margin to equity ratio on the TF portfolio.

Dan G. wrote:
I've been thinking the only way to get a reasonable number for VAR in a TF system would be to run a monte carlo simulation with proper assumptions on a fully loaded portfolio and calc the VAR. Once this number is in hand, then adjust VAR for the current level of 'loadedness' of the portfolio.
The VaR numbers I have seen on TF portfolios did use this method, historical simulation, which is one of several ways of calculating VaR. I believe the VaR literature suggest different lookback window lengths for which to calculate VaR. 30 or 90 days are the typical lookback lengths I have seen used. The lookback period is used to simulate the market volatility and correlations which VaR incorporates. Since positions and degree of leverage can change considerably in most TF portfolios, so will a 95% (or 99%, etc.) VaR. The VaR depends to a large extent on the long/short offsets among markets and sectors as well as how volatile markets (with current positions in the portfolio) are in the lookback period.

The original Turtle System incorporated these concepts in a different way but did utilize these concepts and that was several years before the first VaR literature was published.

JimR
Full Member
Full Member
Posts: 21
Joined: Sun Feb 27, 2005 5:03 pm
Location: New Jersey

Post by JimR » Wed Mar 23, 2005 5:14 pm

wonkabar wrote:My interest in VaR is along the lines of being curious to what the number would be. I know banks and funds tend to use things like this and think it would be interesting to see what the numbers say. It is just a tool like anything else and people who over rely on the number generated and lose focus as to how and why their portfolios are being constructed will inevitably be in for a shock at some point. The only sure thing is that we do not know what our true risk is and there is no way of measuring it since we can not predict the future.
I tend to agree with this view, particularly when Taleb used the analogy of a pilot using a faulty altimeter to summarize VaR theory:

This is an excerpt from the Derivatives Strategy interview titled "The World According to Nassim Taleb" which can be dowloaded from his site:

"DS: Proponents of VAR will argue that it has its shortcomings but it's better than what you had before.

NT: That's completely wrong. It's not better than what you had because you are relying on something with false confidence and running larger positions than you would have otherwise. You're worse off relying on misleading information than on not having any information at all. If you give a pilot an altimeter that is sometimes defective he will crash the plane. Give him nothing and he will look out the window."

This is a very intelligent analogy, one the nobel prize winners at LTCM might have pondered. If you had to compare trading to another activity, flying would probably a fairly close comparison.

JG
Contributing Member
Contributing Member
Posts: 6
Joined: Fri Jan 07, 2005 5:48 pm

Post by JG » Thu Mar 24, 2005 9:45 am

If VaR is assumed to be the max possible loss over the VaR time window and leverage is routinely applied as such, I agree with the Taleb argument.

Post Reply