Good Morning, I have been asked to put together an asset allocation study for an endowment that is currently investing a significant % of their assets in one hedge fund of funds and the remainder in more standard asset classes. I was thinking of illustrating my concerns with this strategy (assuming it is inefficient) with an exhibit that illustrates three points: 1. An efficient frontier using standard asset classes 2. An efficient frontier using the asset classes currently in the portfolio 3. Where their portfolio currently lies in relation to both efficient frontiers given their current allocation First problem I am having is generating the efficient frontier using multiple asset classes. I can only assume that I would have to use an iterative process to go through all combinations of the different asset classes to find the combinations that give the highest sharp ratios. If someone has some experience with this I would really appreciate some guidance. Second problem I am having is that the hedge fund of funds has only been around since the beginning of 2008 so I really won’t have enough data to determine long term return/SD/correlations against the rest of the asset classes. Does anyone have any ideas on how to get over that hurdle? Third, I want to be able to propose an alternative allocation that would lie on whichever efficient frontier is higher (I would assume the one that includes the alternative asset class). Does anyone have any experience with generating corner portfolios given return/sd/correlations of each asset class? Finally the client has a $ goal that they would like to reach by a certain time and I would like to run a monte carlo simulation to determine how likely they are to meet this goal taking into account expected donations into the fund and expenditures out of the fund. A quick google search gives me a number of hits on running monte carlo in excel and I can definately go through each of these but I was just wondering if anyone has a specific link that they have used and found to be useful. I appreciate any guidance you can give me.

the hedge fund of funds has only been around since the beginning of 2008 --Use an index --Make estimates of LT risk and return First problem I am having is generating the efficient frontier using multiple asset classes. Third, I want to be able to propose an alternative allocation that would lie on whichever efficient frontier I would like to run a monte carlo simulation --There are a number of software packages that do this for you. Even if you dont have access to one, you may want to consider looking into a free trial.

MPI or Zephyr should be able to do all of those things - not sure if it’s worth the learning curve though. If you go the excel route, you can create a efficient frontier pretty easily using solver. YOu need the inputs though - return, stdev, and correlation. If you’re not interested in building forward-looking capital market assumptions, just use indexes to proxy all of your asset classes and use the historical. For the monte carlo, riskamp is a pretty inexpensive excel add-in (similar to crystal ball)

Thank you both for your suggestions. I am attempting to learn Encorr Optimizer and Zephyr Allocation Advisor today since those applications seem to be able to do what I want to do.

https://www.hedgefundresearch.com/mon_register/index.php?fuse=login&1245174020 Use this: HFRI Fund of Funds Composite Index From the return stream, calculate the historical risk/return (and use that to project your expected return and expected standard deviation based on your future assumptions about the asset class). The problem w/ this approach is that all hedge fund strategies suffer from left tail distribution and high kurtosis: a typical mean-variance approach will definitely underestimate your volatility. A simple adjustment will be to “scale” up your expected standard deviation. I would suggest (if you have the time) to look into the Michaud Resampling approach. It’s a more advance mean-variance technique that provides more stable portfolio allocations as you move along the frontier. The problem w/ the traditional MVO approach is that as you move slightly along the frontier, your allocation to each asset class shifts DRAMATICALLY… For example: if you move from a very conservative portfolio (bottom left of the frontier) to say a midpoint level, your cash/FI allocation can shift from 50% to say 5% easily. Putting constraints will alleviate the problem, but you still suffer from your own modeling biases. I have couple of papers on this, but you can just google him up (one thing I found was: http://www.pionline.com/article/20031222/PRINTSUB/312220715)

Recognizing that hedge fund manager universes violate almost all of the necessary features of a benchmark, I would suggest using HFRI (Hedge Fund Research) Fund of Fund indices for modelling purposes. The time series dates back to 1990 and you can select from an array of strategies. If you have EnCorr with data updates, then you should have it. If not, I can e-mail it to you. EnCorr will also allow you to run a Monte Carlo simulation up to 5000 trials.

Forgot one thing: I would also recommend that you unsmooth the HFRI returns to account for serial correlation. Private real estate, private equity, and to a lesser extent hedge fund of fund returns are serially correlated, which biases volatility downward. You can unsmooth returns using the Fisher-Geltner-Webb unsmoothing calculation: First, you need to determine the correlation coefficient, b, at lag one using autoregression: Rt=a+bRt-1 Rt is return at time t, Rt-1 at time t-1 b represents the serial correlation coefficient at time t Next, unsmooth the return stream using b from above with the FGW calculation: Rt (unsmooth)=(Rt-bRt-1)/(1-b) The return stream above will more accurately reflect underlying asset class volatility.

IA Encorr Optimizer has the option to use a Resampling method but I am finding that I still have to put a ton of constraints in just to get it to include all of the asset classes. For instance, the expectation is that by adding international exposure we should be able to shift the frontier out due to the diversification benefits, but the optimizer keeps leaving international out and when I force international in through a constraint, the frontier actually shifts in (i.e. less return for any given level of risk). This is using return series that go all the way back to the 1970’s. This would suggest that we are not getting a benefit from international exposure and all of these target date funds that include up to 20% international have it wrong. What am I missing here? FYI: I am using DJ Wilshire 5000 to represent the U.S. Equity Market, MSCI World ex. US to represent the international equity market, Barclays Intermediate Credit to represent Fixed Income, and 90 day T-Bills to represent cash/short term.

Are you using historical as your expected returns? Or your own assumptions based on building blocks, Black Litterman, CAPM…? If the former, common period returns constrained by the shortest time series?

djschulz Wrote: ------------------------------------------------------- > Are you using historical as your expected returns? > Or your own assumptions based on building blocks, > Black Litterman, CAPM…? If the former, common > period returns constrained by the shortest time > series? Right now I am just using historical because I figured that would be the easiest way to get my feet wet with the application. I’m using common period returns constrained by the shortest time period. I am only taking returns through December 2007 because for now I am considering 2008 to be an outlier that will skew my long term expections for these asset classes.

I work at a large investment consulting firm and we do our own projections… As a firm, we normally project out our assumptions using: for fixed income: implied forward rates from the yield curve for equities: growth approach (earnings growth + p/e expansion + dividend) risk prem approach (10 yr treasury yields + equity risk prem) all accounting for inflation, earnings beta, nominal growth, p/e expansion etc etc

i would avoid historical returns if possible: take a look at this presentation that zephyr put out. http://www.styleadvisor.com/resources/conference/2004/AllocationADVISOR%202004.pdf

p.s. i have a good short paper that describes the Michaud resampling process if anybody is interested. leave your email here and i’ll send away.

adalfu Wrote: ------------------------------------------------------- > i would avoid historical returns if possible: > > take a look at this presentation that zephyr put > out. > http://www.styleadvisor.com/resources/conference/2 > 004/AllocationADVISOR%202004.pdf I read through the presentation and agree that what is happening to me is exactly what they say will happen if you use historical data. Looks like I will be using risk and return assumptions going forward. Thank you very much for opening my eyes to this. Gotta read the Allocation Advisor manual…182 pages…by thursday so that I can start putting together this report. Good times.

djschulz Wrote: ------------------------------------------------------- > Forgot one thing: I would also recommend that you > unsmooth the HFRI returns to account for serial > correlation. Private real estate, private equity, > and to a lesser extent hedge fund of fund returns > are serially correlated, which biases volatility > downward. You can unsmooth returns using the > Fisher-Geltner-Webb unsmoothing calculation: > > First, you need to determine the correlation > coefficient, b, at lag one using autoregression: > > Rt=a+bRt-1 > > Rt is return at time t, Rt-1 at time t-1 > b represents the serial correlation coefficient at > time t > > Next, unsmooth the return stream using b from > above with the FGW calculation: > > Rt (unsmooth)=(Rt-bRt-1)/(1-b) > > The return stream above will more accurately > reflect underlying asset class volatility. This is a must-read if you want to correct for non-normality of market returns: http://www.jpmorgan.com/pages/jpmorgan/am/ia/research_and_publications/white_papers go to the “complete paper” it should be 40 pages long. djschulz above only addresses correcting serial correlation by unsmoothing and correcting the raw return. there are other things that you can do as well: using EVT to correct for fat tail risk, or simulating correlation breakdown by using copula theory.

adalfu Wrote: > This is a must-read if you want to correct for > non-normality of market returns: > http://www.jpmorgan.com/pages/jpmorgan/am/ia/resea > rch_and_publications/white_papers > > go to the “complete paper” it should be 40 pages > long. > > djschulz above only addresses correcting serial > correlation by unsmoothing and correcting the raw > return. there are other things that you can do as > well: using EVT to correct for fat tail risk, or > simulating correlation breakdown by using copula > theory. Thanks for the link. I read it and came to the realization that I should have paid more attention in my linear algebra class so that I can set up an optimization routine using their method. I keep telling myself that if this company is going to start doing more detailed analysis I am going to have to go take some basic math courses at the local community college to brush up. I haven’t done any serious math in about 10 years.