The following benchmarks have been suggested for setting VaR limits:1. 1-day VaR limit = 5% to 10% of Treasury's revenue budget (before operex) for the year2. 1-day VaR limit = 5% to 10% of the bank's net income budget for the year What other benchmarkes are used for setting VaR limits?

I'm like blown away here. OK. The other limits include 1%, 2%, 3%, pi% of the number written on the wall (I can't tell you which wall that is).You should look at VaR in the context of risk vs. return and what if the shit's gonna happen twice as big as what has happened in the history of the universe already since the Jesus was born, will I survive the blow and will be able to operate in that case? Those should be your limits.

are the worst cases beyond var anyway? if u set var limits based on outliers won't it encourage more risk taking ? u should then have stress limits ?

Since as one of the purposes of capital is to stand ready to absorb trading losses, it might make sense to think about VaR in relation to capital as well. And while it is of course important to make all calculations as useful as possible, remember that firms compute VaR more because they have to (to please the regulators) than they want to.

QuoteOriginally posted by: DavidJNSince as one of the purposes of capital is to stand ready to absorb trading losses, it might make sense to think about VaR in relation to capital as well. And while it is of course important to make all calculations as useful as possible, remember that firms compute VaR more because they have to (to please the regulators) than they want to.Yes, although to be safe, one should set VaR against equity so that that the greater the leverage, the more conservative the limits relative to the total book.The key with VaR is to remember that VaR is not the most you can lose, but it is the least you will lose when things go wrong.

QuoteOriginally posted by: Traden4AlphaQuoteOriginally posted by: DavidJNSince as one of the purposes of capital is to stand ready to absorb trading losses, it might make sense to think about VaR in relation to capital as well. And while it is of course important to make all calculations as useful as possible, remember that firms compute VaR more because they have to (to please the regulators) than they want to.Yes, although to be safe, one should set VaR against equity so that that the greater the leverage, the more conservative the limits relative to the total book.The key with VaR is to remember that VaR is not the most you can lose, but it is the least you will lose when things go wrong.This is a very simple and intuitive idea yet it seems like many Risk managers and CEOs are unable to understand this idea of using VaR as a tool to protect capital. People should set up VaR limits on a proactive manner [like T4A and DavidJN said] as opposed to setting up a stupid limit based on the historical series and/or just using VaR for regulatory purposes. Thinking and defining extreme loss scenarios should also be incorporated in capital requirements. Plenty of people use VaR/stress testing as just a black-box (meaningless) number instead of using it as a proactive risk managament metric to avoid huge capital losses.

Balance the two sides of the coin. The original poster mentioned VaR limit setting at the income perspective while the replies set limits using the risk perspective. The "right" number balances these two perspectives because becoming too strict with your VaR limits will make your budget targets unattainable. VaR is not just a measure of probable loss but aslo probable gain.

QuoteOriginally posted by: jomniVaR is not just a measure of probable loss but aslo probable gain.V@R = "Value At Risk", i.e. by definition it deals with potential losses, not gains.Probably, jomni, you're assuming some particular distribution... Maybe Gaussian?

Yup gaussian. But even if it's not, it is still a rough measure of probable gain. You won't earn much from an instrument with low risk. We must go beyond the traditional definition for it to be useful. You won't get much buy-in from front office if all you do is look at the risk side. A good risk management framework is not only about limiting risks, it's about helping the bank attain their goals.

QuoteOriginally posted by: jomniYup gaussian. But even if it's not, it is still a rough measure of probable gain. You won't earn much from an instrument with low risk. We must go beyond the traditional definition for it to be useful. You won't get much buy-in from front office if all you do is look at the risk side. A good risk management framework is not only about limiting risks, it's about helping the bank attain their goals.VaR is only a rough measure of probable gain under assumptions of a positive expectation for gain and identical higher moments. Two instruments can have identical expected gain and variances/volatilities, but if they have different levels of skew, kurtosis, etc. then they can have different VaR levels. Maybe that's what you mean by "rough measure" but it seems an important point for complex derivatives that can have extremely non-Gaussian outcomes.Also, two instruments may have identical expected gain, variance, and VaR, but one instrument may be extremely "risky" in the sense that it generates a 100% loss with some probability that is less than the chosen VaR threshold. VaR only documents the threshold of tail risk, but does not quantify the magnitude of that tail risk (i.e., the expectation of loss conditional on being in the tail).

Returning to the initial post. Suppose net income budget is 10.0bn, and VaR per day limit is 5% of budget, or 500mn. This scales up to a 1-year VaR limit of 7.9bn (on a 250-trading day year). As such, a 1-year market loss of up to 7.9bn would be normal from a VaR perspective. Even a market loss over 7.9bn would seem normal -- that's a 1 in 100 year-type of 1-year loss. The last time anything like 2008's market losses were incurred was in 1929, that's about 80 years ago.

QuoteOriginally posted by: aarnedimanReturning to the initial post. Suppose net income budget is 10.0bn, and VaR per day limit is 5% of budget, or 500mn. This scales up to a 1-year VaR limit of 7.9bn (on a 250-trading day year). As such, a 1-year market loss of up to 7.9bn would be normal from a VaR perspective. Even a market loss over 7.9bn would seem normal -- that's a 1 in 100 year-type of 1-year loss. The last time anything like 2008's market losses were incurred was in 1929, that's about 80 years ago.Again, you're assuming that the process is a Gaussian random walk, while in fact it's almost certainly not. You're underestimating the risk.

There is a multitude of very real risk not factored into VaR - Liquidity risk springs to mind. If liquidity dries up, it may very well take a lot longer than one day to liquidate your exposure.Be very wary of thinking of Var numbers ito risk limits, rather think of them of minimum expected losses with probaility associated with your confidence level.

QuoteOriginally posted by: VegawizardThere is a multitude of very real risk not factored into VaR - Liquidity risk springs to mind. If liquidity dries up, it may very well take a lot longer than one day to liquidate your exposure.Be very wary of thinking of Var numbers ito risk limits, rather think of them of minimum expected losses with probaility associated with your confidence level.True but VaR is backtested against realised p&l and how many people price in liquidity daily?In any case VaR has been shown to be one of the most stupid, if unfortunately widely relied upon for real decisions, models ever. This is partly for reasons alluded at above re: assumed distributions.

QuoteOriginally posted by: OblezinQuoteOriginally posted by: aarnedimanReturning to the initial post. Suppose net income budget is 10.0bn, and VaR per day limit is 5% of budget, or 500mn. This scales up to a 1-year VaR limit of 7.9bn (on a 250-trading day year). As such, a 1-year market loss of up to 7.9bn would be normal from a VaR perspective. Even a market loss over 7.9bn would seem normal -- that's a 1 in 100 year-type of 1-year loss. The last time anything like 2008's market losses were incurred was in 1929, that's about 80 years ago.Again, you're assuming that the process is a Gaussian random walk, while in fact it's almost certainly not. You're underestimating the risk.Say the Board mandates that trading book loss for the year should be no more than 2.0bn (that's 20% of income budget of 10.0bn) -- 0% probability of loss greater than 2.0bn. This is a conditional probability (conditional on portfolio reallocations). Regardless of price stochastic processes, the 1 year portfolio VaR is a number less than 2.0bn. VaR, on the other hand, assumes a theoretical static portfolio. (a) Is a 0% probability of loss greater than 2.0bn realistically possible / theoretically founded? (b) What is the 1 year VaR that corresponds to the 0% loss probability.