Hacker News Reader

Comment by dsjoerg | Hacker News Reader

that's effectively only ~4 observations then (100/24). i wouldn't want to make a very important life decision based on data from only 4 observations.

with only 4 observations, "common sense" and other forms of first-principles reasoning are much more important than the 4 data points. also one should consider data from other countries and other historical eras.

View on HN

It's not though. There are ~75 24-year windows you can use between 1913 and today. I suspect you will find similar results with almost all of those windows. Obviously past performance is not an indicator of the future, but 75 data points is a pretty decent amount of data.

Following up, I plugged in the data from here into Excel: http://www.multpl.com/s-p-500-historical-prices/table/by-yea...

Some observations:

  -there are 122 24-year windows in the dataset
  -not a single 24-year period had a negative return
  -the lowest average return for a 24-year window was 0.15% in the window from 1874-1897
  -there have been only two 24-year windows since 1876 with average annualized returns less than 2%
  -every window since 1912 has been at least 3%
  -every window since 1968 has been at least 7.1% (the author's assumption)
  -every window since 1971 has been at least 8%

Note that this doesn't take inflation into account, but also does not factor in the returns from reinvested dividends. Those effects would probably approximately cancel each other out, but hard to say.

Those 75 data points are not independent observations, because the windows overlap so much. Remember that using "year" as the granularity is arbitrary anyway; if we had daily, hourly or by-the-minute data going back to 1913, then we could have more "data points" but we would not have any better insight or statistical significance into the original question.

My point is that having 100 years of investment data is much more like having 4 data points than it is like having 75 or 75 *12 (if you cut the years into months). Even though you literally have 75 data points, they are pretty close to a copy-paste of each other; not statistically independent.

Reply on news.ycombinator.com