rpetrocelli:

I find this interesting because it seems odd that a Monte Carlo simulator would change its allocations in a 5 month period.  Any thought on this?  Is the simulator simply performance chasing?

A simulation is extremely sensitive to its inputs; changing the expected return on a fund from 8.2% to 8.1%, or the correlation between two funds from .50 to .48, can lead to shifts of several percentage points in allocations.  And this can happen even if the fund makes no changes in return assumptions; the latest portfolio of a fund has different sector weights and allocations to market segments, so it should be expected to have different returns and correlations.  These changes can even occur for index funds; Total International has much more in emerging markets than it did last year, and all the index funds have much more in energy.

If you do your own simulations, you will find that small changes have essentially no effect on expected returns and risk.  If Financial Engines expects its current portfolio to have a return of 8.50% and a standard deviation of 20.00%, then it is probable that the alternative portfolio has an expected return of 8.49% and a standard deviation of 19.97% given the same assumptions.  No model can have anywhere near that much precision.

rpetrocelli:

At the beginning of the year, Financial Engines suggested the following portfolio for a market risk of 1.25:

• Total Bond 9%
• Healthcare 7%
  
• Total International 20%
• International Growth 9%
• Selected Value 13%
• REIt Index 6%
• International Explorer 6%
• Growth Index 14%
• Cap. Opp. 16%

Today, it recommended the following portfolio (with the changes in bold):

• Total Bond 9%
• Healthcare 7%
  
• Total International 24% 9 (up 4%)
  
• International Growth 8% (down 1%)
  
• Selected Value 12% (down 1%)
  
• REIt Index 6%
• International Explorer 6%
• Growth Index 12% (down 2%)
  
• Cap. Opp. 16%

I find this interesting because it seems odd that a Monte Carlo simulator would change its allocations in a 5 month period.  Any thought on this?  Is the simulator simply performance chasing?

Hi Petro,

The short answer would be, "Yes."

I kinda like what William Bernstein said in "The Intelligent Asset Allocator;" .....next year's efficient frontier will be nowhere near last year's. Anybody who tells you that their portfolio recommendations are "on the efficient frontier" also talks to Elvis and frolics with the Easter Bunny.

Using the Monte Carlo Simulator, you systematically decrease future returns by increasing the percentages you hold in the more recent higher returning assets. The more recent higher returning assets offer the higher valuations; higher valuations equal lower future returns. The S&P 500 in 1999 might offer a good example; precious metals today.

There's nothing wrong with the Efficient Frontier in my view, but what was efficient 5 years ago is still efficient today. It doesn't need to be updated. It appears FE has a new little tool to play with they don't fully understand, huh?

BTW, good to see you have time to post.

Chin

Hi Petro,

As soon as I posted this, I realized I am not considering the other side of the coin. It is possible FE is not looking at only the MCS, but valuations, and adjusting their asset allocation per these valuations.

Not saying this is what they are doing, but without more information, we must consider that possibility.

Chin

Thanks for all the interesting and thoughtful replies.

About every 6 months, I look at Financial Engines.  I then do some web searches to see if anyone can explain to me how it works, other than that it is a Monte Carlo simulator.  I can't figure out the rationale for making changes to a portfolio 6 months later, and I may contact Vanguard to see if I can obtain any information on this.  I'll report back to you if I learn anything.

Petrocelli 

Hi all.  This is a very nice crossover article to the thread on Monte Carlo Simulation (MCS).  It is true that the available documentation on exactly how Financial Engines (FE) works is very sparse.  Based on available information, here is what I can divine.  First, FE uses a type of models called Style Analysis (common, but very different from my own tool, Quantext Portfolio Planner).  In Style Analysis, a fund or portfolio's projected risk and return is regressed against a series of core indices, such as large cap domestic, small cap domestic, large cap international...The idea is that the vast majority of the performance of many portfolios can be explained simply by their expsure to broad indices. 

When you project the performance of a portfolio using Style Analysis and MCS, you have to come up with an expected return and standard deviation for each of the core indices, as well as the correlations.  I believe, but am not sure, that FE must use something smarter than just historical data.  It probably uses some analyst projections for each asset class--but we cannot know for sure.  If the analyst's estimate shift, the optimal portfolio will shift.  OR if the model simply updates pure historical parameters on a rolling basis, the parameters will evolve in time and the optimal portfolio will therfeot also evolve in time.  This is not a problem in my opinion.  Better to have this approach than purely static parameters for all time--as long as the process for updating parameters is solid.  I have never been able to dig up any meaningful technical details on how they do this, so I remain in the dark. 

A problem that I have with Style Analysis is that it does not work well for assets in the portfolio outside of the core set of indices.  Bill Sharpe himself wrote about this problem with utilities.  Utilities are not well represented in his standard Style Analysis model because they do not map well to broad domestic stock indices (because of their low Beta) and special characteristics as an asset class.  The same is true, in general, for commodities, precious metals, etc.  Now, perhaps FE has added indices for these, but this was not the case when last I checked.

Finally, my main issue with FE is that they don't generate much in the way of validation.  I would love to see a paper showing how their modeled optimal allocations have evolved in time and whether they have outperformed a simple benchmark like a mix of VFINX and a bond fund.  In other words: I'd like to see some data to support the idea that their model generated portfolios are better than something really simple minded.

I have looked in vain for supporting information on the FE model beyond the totally simplistic stuff I can find on their website.  Please post if you find such things.

Regards,

Geoff