On portfolio diversification

Diversification: How useful is it really?

By mak ·

I ran 20,000 equal-weight S&P 500 portfolio simulations across 100 randomly selected historical windows since 2005:

 - 50 five-year periods and 50 ten-year periods
 - portfolios of 5, 10, 15, 20, and 25 randomly selected stocks so even monkeys can benefit
 - Buy-and-hold, plus monthly, quarterly, and annual rebalancing
 - Point-in-time S&P 500 membership

What we can see: marginal returns for diversification are quick and diminishing.

 A few takeaways:

 1. Going from 5 stocks to 10–15 stocks did most of the diversification work. Volatility fell meaningfully; gains beyond that were smaller.

 2. Rebalancing generally improved median returns. In the 10-year sample, 25 stocks returned 207.3% without rebalancing versus 239.6% with quarterly rebalancing.

 3. Rebalancing was not a risk-reduction button. It often came with slightly higher volatility and deeper drawdowns than simply letting winners run.

diversification-1.pngThese Monte Carlo simulation paths goes beyond to see beyond the median. While diversification immediately reduces median annual volatility, we can observe that the 5 stock portfolios are much more “bottom heavy”, while the they share similar 90th percentile paths. so - even if you want to bet big and hope to win a lot, chances are focusing on 5 winners is not going to improve your chances that much better. This is especially true if you like to rebalance your stocks every once in a while.diversification-2.png


This is why I try to limit my individual stock exposure to somewhere between 10-20 stocks at a time. Not only does it optimize on the idiosyncratic risk/return trade off, it also helps with mental capacity of keeping track of so many investment theses at once.

One caveat though: to truly benefit from diversifcation, your basket of individual stocks must be relatively different. If you’re holding 10 stocks and 5 of them are energy producers, expect higher annual volatility than a basket of less correlated stocks.