• Cover's universal portfolio maximizes the wealth growth rate
  • Markowitz's mean-variance model minimizes portfolio variance

Both allocate assets based on historical returns.

How do these two models perform against one another (assuming for Markowitz we use the global minimum variance portfolio by default). How does the universal portfolio compare against the equally-weighted portfolio that is known to outperform Markowitz sometimes? Does the universal portfolio provide portfolio weights in-sample that hold up well out-of-sample compared to the minimum-variance portfolio?


I personally think that the most appropriate answer is: it depends. Specifically what is "Markowitz's mean-variance model"?

If we consider the more general definition it is an optimization framework and the portfolio is obtained by optimizing according to an objective function and it is expression of the estimated mean/return component and variance/risk component.

First of all the point is how one estimates the inputs (which assumptions, models, statistical techniques etc.) and secondly how one defines performance.

If we only consider the vanilla (no constraints etc.) minimum variance portfolio and vanilla growth optimal portfolio and supposedly we compute the variance component equally all becomes a question of how one estimates the mean component required to obtain the second portfolio.

Finally if we suppose that the estimated mean component is quite close to one's expectation and performance is intended "stability" or "Sharpe-ratio" it happens that the growth optimal portfolio is less diversified and riskier/more volatile (at the same time it compound the invested capital faster thus delivering a higher terminal wealth).

Both Samuelson (1971) and Markowitz (1976) implied that many investors are willing to sacrifice long-term return in exchange for short-term stability.

  • $\begingroup$ the estimated mean is often far from what we expected, so which of the two portfolios is better then $\endgroup$ – develarist Sep 20 '20 at 15:01
  • $\begingroup$ The estimated mean component in m.v. optimisation is the vector of expected returns (it is not intended as statistical mean which has been proven to be a poor estimator in this framework). Whatever model one applies if it does not provides good (let it be read as "close to one's expectation") of the expected returns probably it is more rational to consider only minimum variance portfolio simply because there is no need to estimate the mean component. $\endgroup$ – Nipper Sep 20 '20 at 16:43
  • $\begingroup$ thats why the question only discussed the min var portfolio, not the sharpe portfolio. how does your answer change w.r.t. the min var portfolio? $\endgroup$ – develarist Sep 20 '20 at 17:14
  • $\begingroup$ Could you please explain me where I referred to the sharp portfolio (which is also called “max sharpe” or “tangency” portfolio)? $\endgroup$ – Nipper Sep 20 '20 at 20:35
  • $\begingroup$ your 5th paragraph $\endgroup$ – develarist Sep 20 '20 at 21:17

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.