Asset Allocation and MVO

This method can be sensitive to how narrowly one defines an asset class. Are U.S. stocks one asset class or

4 ((large/small cap) * (value/growth))? What asset classes did you use?

What was the turnover using MVO and rebalancing? This would affect the tax efficiency of the strategies if they were not implemented in a tax- deferred account. What periodicity of returns (daily, weekly, monthly) were used to compute expected returns and correlations?

I am asking :). Imposing constraints subjectively, knowing how the asset classes have performed over the full sample, may be introducing hindsight and biasing the results upward.

I could do the math myself, but I think the results are more usefully presented as annualized returns.

The comparison is more meaningful if you constrain the optimized portfolio to have the same volatility as the unoptimized one. What were the volatilities of the two approaches?

Stocks have averaged higher returns than bonds, so the MVO could outperform an equally-weighted portfolio (EWP) if it consistently allocated more to stocks than the EWP did. You could compare the MVO portfolio to one that had the same AVERAGE exposures to the asset classes, to see if the MVO dynamic weights added value.

Thanks again for posting your results. A similar project is on my to- do list.

======================================= MODERATOR'S COMMENT: Posters to this thread should relate comments to general financial planning.

"This method can be sensitive to how narrowly one defines an asset class. Are U.S. stocks one asset class or

4 ((large/small cap) * (value/growth))? What asset classes did you use?"

The eight asset classes were: Large-Cap Growth, Mid-Cap Growth, Small- Cap Growth, Large-Cap Value, Mid-Cap Value, Small-Cap Value, International, and REITs. For investments, iShares are now used for all the asset classes.

"What was the turnover using MVO and rebalancing? This would affect the tax efficiency of the strategies if they were not implemented in a tax- deferred account. What periodicity of returns (daily, weekly, monthly) were used to compute expected returns and correlations?"

The "MVO Portfolio" was rebalanced once a year. Assume one used TD Ameritrade with commissions of $7.00 per trade (that is what I pay), if one made a round trip for each asset class, commissions would be 8 x 2 x $7.00 or $112.00 per year. It is not unusual for an asset class not to require a percentage change, but the dollar amount would change so I would assume one would pay over one hundred dollars in commissions each year.

I used five-years of data (Evensky recommendation in his book, "Wealth Management.") to calculate the arithmetic, geometric, standard deviation, and correlations. Portfolio returns were calculated on an annual basis. I need to go back over my SS to make sure I did not have errors. The dollar difference seems too good to be true.

Physlab PS I hope this message gets through as I have had a difficult time getting my mail to "stick."

======================================= MODERATOR'S COMMENT: Posters to this thread should relate comments to general financial planning.

I am asking :). Imposing constraints subjectively, knowing how the asset classes have performed over the full sample, may be introducing hindsight and biasing the results upward.

"I could do the math myself, but I think the results are more usefully presented as annualized returns."

I could post the annual returns if that would be helpful.

"The comparison is more meaningful if you constrain the optimized portfolio to have the same volatility as the unoptimized one. What were the volatilities of the two approaches?"

I did make one run were all asset classes contained the same percentage or 12.5% of the total portfolio. In my personal portfolios, I use a 30% upper and lower limit to identify when an asset class is out of balance. For my first experiment, I use constraints of 60% on the low side and 150% on the up side. For example, if an asset class was assigned 10% of the portfolio, then the lower constraint was set at .06 and the upper constraint at 0.15. This means I want at least 6% invested in this asset class but no more than 15%.

"Stocks have averaged higher returns than bonds, so the MVO could outperform an equally-weighted portfolio (EWP) if it consistently allocated more to stocks than the EWP did. You could compare the MVO portfolio to one that had the same AVERAGE exposures to the asset classes, to see if the MVO dynamic weights added value."

I did not include bonds in this analysis as I do not use bonds. Bonds are a drag on a portfolio. Even though I am retired, I stay away from bonds as I derive income from social security and pensions. I consider those to be "bond equivalent" holdings, even though I do not control them. I picked up this idea from Bernstein's second book, "The Four Pillars of Investing."

My MVO Portfolio returns were compared with identical ETFs, only the asset allocation percentages were different between the two portfolios.

"Thanks again for posting your results. A similar project is on my to- do list."

Please post your results when you complete this project. I now need to go back and do a desk check on my SS formulas to make sure I did not include errors. An Internet friend is checking my SS for errors.

I appreciate your questions as they help me rethink what I am doing.

Physlab

"I am asking :). Imposing constraints subjectively, knowing how the asset classes have performed over the full sample, may be introducing hindsight and biasing the results upward."

Here are my asset allocations when starting this experiment.

Large-Cap Growth = 10% (IVW) Mid-Cap Growth = 10% (IJK) Small-Cap Growth = 5% (IJT)

Large-Cap Value = 10% (IVE) Mid-Cap Value = 15% (IJJ) Small-Cap Value = 15% (IJS)

International = 25% (EFA) REITs = 10% (ICF)

I am working to break the international into Int'l and Emerging Markets (EEM) where the emerging markets will include China (FXI) and other ETFs. The emerging markets asset class will comprise 10% of the portfolio.

The logic for the above percentages is based on Fama & French's 1992 article and my personal experience when I set up a passive portfolio back in late 2000.

Physlab

Physlab,

Thats a pretty good asset allocation you got there! By all means continue your research, but I don't think you are going to glean that much more from breaking down your asset classes further.

The asset allocation model I use essentially IS a more broken down version of yours. I plugged your AA into the professional planning software that my firm uses and it lies on the same frontier as ours. IOW, even though we divided our asset classes even further we still ended up with an almost identical expected return, std dev., and sharpe ratio. As a matter of fact, yours slightly outperforms ours.

Your portfolio generated a 10.45% expected return, 11.64% std dev., and 0.47 sharpe ratio while ours generated a 10.46% ER, 11.96% SD, and

0.46 sharpe. Of course this is all based on the software's assumptions of variables, but the assumptions were held constant when comparing the two portfolios (apples to apples and all).

My AA:

Cash = 2% LgCap Value = IVE = 12% LgCap Gwt = IVW = 15% MidCap Value = IWS = 7% MidCap Gwt = IWP = 10% SmCap Val = IJS = 9% SmCap Gwt = IJT = 10% Int'l = EFA = 20% Ntrl Resource = VDE = 4% Emerging equity = EEM = 7% Real Estate = RWR = 4%

You may even want to toy with consolidating your asset classes. The software calculates that one will outperform both of the above allocations with the following portfolio: (10.5% ER, 11.70% StdDev, and 0.48 Sharpe)

IWF = 25% IWD = 25% IWR = 18% IWO = 6% IWN = 6% EFA = 20%

***insert historical, hypothetical, yada yada disclaimers here***

Using 5 years of historical data to estimate volatilities and correlations is plausible, but using only annual data throws away too much information -- the standard errors of estimates using 5 observations will be too high. It would be better to use monthly or daily returns, as discussed in a paper by Stephen Figlewski that I previously mentioned here

.

Forecasting future returns using the average returns over the last 5 years is poor -- one should use a much longer time period or have a model based on valuation. There have been periods where the 5-year returns on stocks were close to zero or above 20% -- it has been risky to extrapolate from such periods.

here

.

How many months do you consider to be a minimum? Is there a maximum number of months you recommend?

I am in the process of adding months of data to my MVOPlus data source, as well as other asset classes.

Physlab