Portfolio Composition 2 part question

Visibility: Public
asked Feb 04, 2015 12:37 AM by
Pierre Guenette gravatar image

In the Portfolio composition blurb, it states that you use historical prices of the security itself and in relation to the others. I was curious as to how far back do you go in constructing your variance-covariance matrix (i.e., do you use an arbitrary number like 20 years or do you go as far back as there is data)?

Also, how do you adjust for possible skewing of the data after a security has had a massive run. For example, in the last few years bonds have had an amazing run (although there isn't much room to cut interest rates further) that I would imagine a Sharpe ratio optimizing model might allocate a greater portion of the portfolio to that asset class given it's high returns and low risk profile.

Edit Retag Flag offensive Close Delete

1 answer

sort by ยป oldest newest most voted
answered Feb 04, 2015 03:18 PM by
Jin Won Choi gravatar image Administrator

Hi Pierre,

The "historical return" on the model portfolios page is actually synthetically constructed using benchmarks. For example, U.S. stocks had returned 9% over the past 100 years, so the U.S. stock component of the portfolios (currently CAPE) is assigned a "historical return" of 9%.

Let me explain why I chose this approach. As much as I'd like to give a forecast of what I think the portfolios will return over the long run, I can't do so because of regulatory reasons. I chose to display a synthetic historical return instead because I recognize that most people set their expectations based on such a number. If I were to display actual historical performances of each portfolio, the rate of return would be far higher.

In the future, I plan on introducing a long term stock market forecasting model. I'm hoping that this model will do a better job of setting expectations for the performance of the portfolios, rather than the "historical returns". Because it's a model, I think I can avoid regulatory issues since I'm not personally forecasting anything. I'll just be letting the model output some data.

I hope that answers your question, Jin

Edit Flag offensive Delete Link
Please log in to post an answer.
Web Analytics