What Mean-Variance Portfolio Weights Tell Investors
Summary
The document explains that portfolio optimization determines asset weights under a chosen objective; it does not automatically identify a small set of stocks expected to outperform. In mean-variance optimization, the inputs include expected returns, variances, and correlations. With a no-short-selling constraint, an asset that contributes poorly to the objective may receive a zero weight. Allowing short sales can instead produce negative weights, while the weights still sum to the specified total.
The answer frames optimization as a way to choose a portfolio’s risk-return tradeoff, often expressed through the Sharpe ratio, rather than as a standalone stock-picking rule. It says that historical returns are commonly used to estimate inputs, but they need not represent future performance; an investor’s own return expectations could be substituted. The theoretical case depends on assumptions including risk aversion, reasonable estimates, and normally distributed returns. Departures from these assumptions, especially heavy tails, limit how reliably the optimized allocation represents an investor’s best choice.
Key ideas
- Mean-variance optimization allocates weights using expected returns, variances, and correlations.
- A long-only constraint can assign zero weight to assets that do not improve the chosen objective.
- Allowing short sales permits negative weights while portfolio weights can still sum to the target total.
- Optimization specifies portfolio composition under assumptions; it does not by itself provide a stock-selection signal.
- Historical return estimates may not represent future performance, and distributional assumptions affect the result.
Tags
Full text
# What does an optimized portfolio really tell us? # What does an optimized portfolio really tell us? I am very new to this field, and have very recently started doing some self study on this topic. After reading some papers and reproducing some of the results in them, I am not very clear about what objective exactly is fulfilled by doing portfolio optimization. I assume that having a minimum variance portfolio is of interest or relevance for investors because it tells them objectively which stocks to buy and which ones to sell, is it not? If that is not the case, then could you please explain why would an optimized portfolio be of interest to investors? Also, if we say that investors should invest in stocks according to the optimized portfolio weights, does it mean that investors should invest in all the stocks anyway – and this method just gives the proportion of each to be invested in? If yes, it is a bit confusing, because in real life, although I myself have not traded, I assume that an investor would typically buy or sell only a small subset of stocks at any given time. Is this not true? So, the more meaningful problem to solve would be, in my mind, to find out which stocks to buy or/and sell at any given time, based on the studied stock returns, is it not? EDIT: For example, we can use www.portfoliovisualizer.com to check which assets to invest in (according to whatever theory they are using): The algorithm is not giving as output a portfolio of all the investments that the investor initially had. It is giving a subset of them (I assume the ones most likely to make profit). But portfolio management theory doesn't seem to talk about subsets of assets. Which one is correct, or better? ## Answer by Ana (score 2) https://quant.stackexchange.com/a/27518 It means you should buy the 4 stocks. The model you are using seems to restrict short selling, i.e. by removing this restriction you could get negative weights on certain assets with all assets adding up to 100%. Re your question on limiting your asset selection to stocks that are expected to outperform the population: this has to do with your assumptions behind expected returns. Remember, that the important inputs for your mean variance optimization are expected return, variance and correlations between the assets. Generally historical data is used for this purpose, but this doesn't necessarily reflect the expected future performance of the assets you are about to invest in. You might just as well substitute the historical average returns with your personal expectation of the returns, and if you e.g. expect certain assets to yield negative returns with positive correlation to your total portfolio, then obviously it doesn't really matter whether you include that stock in the selection or not, if short selling restriction is imposed then 0% weight will be allocated to that stock. Bottom line: PT maximizes your Sharpe, that is the expected excess return of your total portfolio divided by the expected total variance. Assuming that you are risk averse, the returns are normally distributed (i.e. heavy fat tails don't limit the extent to which you can lever your portfolio, and hence Sharpe is reasonable measure of the risk-return trade off) and the inputs are reasonable, then theoretically PT should yield the optimal composition of your portfolio. Relax these assumptions, and MPT runs in to some of the problems that you are wondering about.
Shown in full with attribution under the source's licence. Licence: CC BY-SA 4.0 (Stack Exchange)
This summary was written by Stratmill's research agent from the original; it is not a copy of the source.