simulating hedge funds

tatana · July 14th, 2004, 6:56 am

Hello,I want to simulate a hedge fund but I do not know which is the distribution that best fit it. Could anybody help me whith this?Does anybody know an article where I can find information about this?Thanks.

Kvadrik · July 14th, 2004, 10:20 am

I think that simulation and back testing formalized daytrading strategies will help you. Can you describe your problem by example?

secondMan · July 14th, 2004, 11:12 am

i think this is going to be a strange thread ...

kritchey · July 14th, 2004, 5:16 pm

You need to provide much more detail about what the hedge fund's strategy is. Unfortunately, a broad strategy category would be not be that helpful as the details of the strategy- anomaly being pursued, degree of hedging, stop losses used- vary considerably even within strategies.

exotiq · July 16th, 2004, 4:39 pm

I saw a talk about one FoF manager using a two dimensional distribution: one describing the distribution of fund returns and a second describing the distribution of return distributions reflecting the uncertainty about whether the fund follows a particular strategy. Without much transparency it is difficult to get past these Rumsfeldian "uncertain uncertainties" models and differentiate alpha from schmalpha, but that's why someone is paying you for the model. Sometimes information about the style of the fund and any quotes you might get on fund derivatives might help...

secondMan · July 16th, 2004, 7:25 pm

get historic index data. add volatility. do not trust the outcome, except for academic purposes. there are so many ifs and whens and bias here and there. figures will strongly misguide you. i would not recommend taking these numbers for pricing options on pools or cdos of hedge funds or that kind of thing.

Aaron · July 17th, 2004, 1:41 pm

Hedge fund return distributions exhibit autocorrelation, negative skewness and kurtosis. Since you are simulating rather than solving equations, the simplest approach is to use two Normal distributions, one with positive mean and moderate standard deviation, and the other with large negative mean and large standard deviation. Each time step there is a transition probability, say 0.05 of going from the first to the second set of parameters and 0.5 of going from the second back to the first.I would capture the autocorrelation by simulating an independent series and taking a moving average, maybe an exponentially weighted one with a decay of 0.8 or so.Anyway, you can tweak the parameters, the two means and standard deviations, the two transition probabilities and the decay factor, to atch the empirical behavior of the funds you are studying. For most purposes you're not interested in the precise dynamics of the returns. If you're constructing a large portfolio of funds for long-term investment purposes, capturing the first few moments approximately correctly is sufficient.One big advantage of this approach is you can account for survivorship bias. You can simulate 1,000 funds for five years, and terminate all funds with drawdowns of more than 50% (for example). Then you adjust the parameters so the remaining funds match your actual data.

Xclr8tr · July 18th, 2004, 2:03 am

QuoteOriginally posted by: tatanaHello,I want to simulate a hedge fund but I do not know which is the distribution that best fit it. Could anybody help me whith this?Does anybody know an article where I can find information about this?Thanks.tatana,Well, you haven't made it very clear what exactly you are trying to do. What hedge fund? Anyways, I ran across an article not too long ago that mentioned that using a lookback option on the markets that these funds are trading would potentially provide you with a better baseline reading on the minimum that a hedgefund should be achieving in the markets they are trading. I thought i saved it but can't find it right now. I'm not even sure that this will help. I believe they used straddles instead of an actual lookback to simulate and rolled them each time a new strike was hit.

Aaron · July 18th, 2004, 1:50 pm

Here is the paper by William Fung and David Hsieh.

tatana · July 19th, 2004, 7:03 am

Thank you for your help.

Kvadrik · July 19th, 2004, 10:49 am

Such distribution may be considered as the total of independent random variables with lognormal distributions. Some results can be calculated from CAPM.

secondMan · July 19th, 2004, 11:42 am

initial poster, would you mind sharing the purpose of this endeavour?peace

britannia · July 23rd, 2004, 6:08 pm

goto www. alternativesoft.com

Kvadrik · July 28th, 2004, 8:59 am

May be it will be interesting for hedge funds?