Risks of Risk Management

[aaron]

This is another common confusion. Static limits, such as position limits, are not meant to be broken. You might raise the limit but (a) you do it before it is broken, not afterwards and (b) you do it when things are going well, and risk is low.



VaR, and other dynamic limits (i.e. limits on values that can change with market or parameter moves, without trading) are meant to be broken. You set the limit. As long as exposure stays under limit, you pay more attention to other things. If exposure is consistently under limit, you lower the limit. Not because you want to reduce risk, but because you want an alert if risk increases.



When VaR goes over limit, you pay more attention. You decide whether or not you are comfortable with the new amount of risk, or if you want it to go up or down. But even if  you decide you want it to go down, you don't force it by telling the trader or desk to get under the VaR limit. Reducing risk can often mean increasing VaR, at least in the short term. In any case, once you're in the abnormal market, VaR doesn't tell you much anyway.



An easy way to demonstrate some of this is to consider running an S&P500 index fund over the last ten years. One guy runs it to a constant VaR, shifting cash into treasuries when VIX goes up (or historical volatility if you prefer) and moving it back to the indes when volatility goes down. The other runs a constant index exposure with the same average investment. The "risk-managed" guy will have lower return and more risk than the guy who ignores risk. From either a risk or return standpoint, the best time to be in the market is when volatility is high. You will find this is true for the large majority of markets and time periods. So managing to fixed risk limits is counterproductive (an apparently similar technique of managing to a target volatility is different since it attempts to control the realized volatility over a period rather than the instantaneous volatility at all times).

[Crassus]

Decent managers are oft hampered by poor/inefficient systems, which preclude any value-added.



A risk manager's bonus is not linked to any measurable. It's based on politics. In fact his real incentive is not to approve any trades, for there is only downside for him when a trade goes wrong.



I infer that risk mgmt is not adding value and presents an inefficient constraint to doing business.



Even the mission statements such as 'we protect the shareholder' are rubbish.





I may pick up the rant later, but that's it for now.



edit:  amusing, but best left for the martini session

[sfca]

I don't think its enough to say most risk managers are idiots.  I do agree that many in risk management are idiots, but thats not my point.  There is something about the process that is dysfunctional.  If you read material on LTCM they did have very bright people running the portfolios and yet they designed risk management systems which failed.  Risk management takes a broad range of skills including knowledge and experience in computer systems, finance, valuation, history, economics, foreign relations, office politics, ability to run meetings, play golf with executives as required, and so on but also most importantly the skill of common sense.   Thats a lot to ask for.  Too much.  Something has to give.  And on top of all that the results of risk management are in probabalistic terms so results are difficult to quantify.  And senior management just does not have the skills much of the time to judge the quality.  And then, the pay is better in other areas so why stay in risk management at all.  The process just does not result in the best solution any more than our political process really puts forth the best two presidential candidates out the the couple of hundred million possibilities.

[doobs]

Crassus,



When I worked at a Bulge Bracket, I had exactly the same view but after i moved to a different setting I realized that the risk folks are probably some of the smartest people around.

[dgn2]

I was a risk manager for a while and I left because the incentives to stay just were not there. That said anyone who thinks that risk management at a dealer is actually about managing anything other than the process of minimizing regulatory capital is a fool. If your guys in risk are clowns, the regulator can require you to hold more capital which reduces your return on capital. That hits the bottom line. This is about leverage and the risk group is there to maximize your allowable leverage. If you think a dealer can operate profitably without these people you are sadly mistaken. I am tired of hearing people slag the back and middle office. These people work like slaves a lot of the time. Without good operations a trading venture is at a severe disadvantage. It is foolish to ignore the cost side of the business and focus entirely on revenue. If I seem hostile about this I am. Years ago our group reduced the dealer’s capital requirements by a billion dollars. That is a billion dollars of capital more to expand the business. If you think that adds no value you are a fool. Low costs are a hard edge and back office and middle office people are the ones that secure these edges.

[Crassus]

edit:  amusing, but best left for the martini session

[Crassus]

now that i've blown my load, i feel better.



Crassus is back in the saddle... gimme some C++, biatch.

[kr]

VaR is always going to look bad applied to a specific asset class. One of its huge advantages is it can be aggregated across asset classes. If all you do is trade CDO's, there are far better risk measures. But if you want to use one pool of risk capital to support lots of different trading and lending businesses, VaR makes more sense.



-----------------------



I'm still struggling with this - it just feels a bit too much like these terrorism alert scales. If you get to the red level, does it mean you should stay home? There are market volatility indices - ok, the sailing is getting rougher, but what is the actionable thing here? Find a new business? In an aggregated form, does it really tell management that the risks of being in their current portfolio of businesses has gone up? I mean, that much can be had from the morning news. I really don't see how it would lead to a different strategy than meeting with the desk heads and working out a plan to change the course of the battleship, unless you don't trust the desk heads to give you an unbiased view of their business. Is that what it's all about?

[apine]

on the one hand, all var is trying to do is put a number on what should happen under the same conditions that held yesterday. i think that is useful information. but does it really give me a measure of my risk? no. i mean, if you think about, all these imploded cdo positions actually have LESS risk now than they used to. right, risk has gotta be less at 40 cents on the dollar than 95 (i am just making numbers up) and after there has already been some liquidation. and maybe that is not the case for cdo's due to default, but you get the idea. one could argue that there is more information now about bear stearns than before that changes the view, but really everybody knew that the financial system is based on leverage. there is no more risk in the system today than there was in january. it is just that now people are pretending to be more aware of it -- and pricing it more appropriately.



i have seen summaries of economic experiments conducted at caltech, i think. they put some small amount of real money on the line (in terms of winnings, not losses). then students go to work trading a security. the security is a wasting asset, like an oil well with known reserves. and the price process looks nothing like a discounted value of the well. if i recall correctly, it starts out close and then runs away in a frenzy of buying as participants only way to make money is either to be long and appreciating or clipping the dividends. and then, like two turns before the well is dry, the asset price craters. sound familiar?



was the risk any different when the asset was climbing up at nosebleed prices or after it starting into freefall? i think the real measure of risk was its departure from value -- but that is more of a trading issue than risk management.

that is probably why so many people are closet indexers.

[dgn2]

apine, since the VaR works in terms of percentage shocks converted to dollars, as the price drops the dollar VaR can actually get lower, although the percentage loss will increase with volatility. Whether the VaR even rises or not depends on the historical window used by the VaR system. One of the things I don't like about VaR (and there is a lot not to like) is that from my perspective a long winning streak increases your potential drawdown even if your one day loss stays the same or even declines. It is easiest to illustrate this concept.



 



[img]/User%20Files/41/clip_image002.gif[/img]



[img]/User%20Files/41/maxDrawdownSlice.GIF[/img]



 



The first set of pictures depicts an equity curve at different percentiles, and the second set of pictures illustrates the associated maximum peak-to-valley drawdown (MPVD). The last larger picture depicts the actual drawdown, the maximum peak-to-valley drawdown, the median expected MPVD, and the MPVD expected tail bounds. Notice that our expectation for the MPVD always increases as we do more trades. In contrast, ETL and VaR level off as on does more trades. The difference between the two measures is sort of related to what apine mentioned above. With the drawdown measure we see that the longer we trade the greater the risk. With the one-day ETL and VaR the number does not grow unless our position size increases or the market gets choppier. The risk is always there and it grows with the number of trades. Our risk of larger draws increases with time in practice even with a strategy possessing a stable edge. If we have a good streak sooner or later we will see a large draw if we do this strategy over and over.



Another one of the problems with VaR is that it focuses on a confidence level that just isn't realistic. We can't be 99% sure of anything. 80% seems more reasonable to me. In fact, I prefer to look at the entire P&L, drawdown, and MPVD distributions.



VaR models are like option pricing models in the sense that if you know them well you can use them to simplify your world. They help one do thought experiments but they can't be used in practice without a feel for the market and that means you need experience. You would no more try to manage with VaR than you would manage a simple options book entirely with the textbook prescriptions of Black-Scholes. People working in a specific part of the business tend to need very different tools than someone overseeing how all of the businesses work together. Most traders I know don't have a lot of intuition about the entire portfolio aggregating all of the institution's businesses and most risk managers I know don't have a lot of intuition about the details of a single book. This has nothing to do with their respective levels of skill or intelligence, but more to do with what is required of them.