The Random Stock Portfolio Experiment

One idea that many investment thinkers have tossed around for many years, even long before the publication of Burton Malkiel’s classic but controversial “A Random Walk Down Wall Street”, is the idea of constructing stock portfolios by pure random selection. The idea has been tested and narrated many ways, with perhaps the most famous image being one of a monkey throwing darts at the Wall Street Journal. As this topic came up recently again on Twitter, I decided it would be worth taking the few minutes to run through and set up a series of randomly generated stock portfolios on the Folio platform, allocating $100,000 of play money (no real money for now) to each, and then plan to check how these portfolios perform relative to their benchmarks over the next few months and years. I call this my “Random Stock Portfolio” experiment.

If you are DIY investor, one BIG reason I would recommend trying this exercise yourself is practice running through the many steps involved in defining investment rules and constructing an executable portfolio, even if the investment rules in this case involve running a random number generator. In this case, the rules and process I chose to go with are as follows:

  • Universe: Russell 3000 index of US stocks (of which 2,863 were able to pass my round 1 check of “clean data” / tradable tickers)
  • Each “Random Stock Portfolio” would be a sample of 30 stocks (about 1% of the names, not 1% of the index weight) chosen at random from the universe, then equally weighted
  • I run the random sampling by sorting the index components by the values of an Excel rand() function, and then selecting sequential blocks of 30
  • I am running 5 of these random stock portfolios at the same time, to get some sample of the variance of what five similar random samples might look like

I would expect these portfolios to have a very strong tilt towards small caps and micro caps, as each of the ~3,000 names is equally likely to be selected, and although each represents 1% of the names in the index, I would expect on average these portfolios to each sample <0.1% of the base index. The benchmark for all five of these portfolios would be the Russell 3000 itself, and I will be keen to see if there are any surprising updates from this experiment over the first 6-12 months.