Reading The Numbers
I listened to a
great EconTalk podcast
last week about the dangers of using small samples, data snooping, survivor bias, and other more technical statistic issues when proving theories or theses. I'll admit, it's not a topic that turns most people's cranks, but I found it interesting because of its relevance to the investment industry.
Small sample sizes
We've all heard politicians, economists, or even friends and colleagues say something to the effect of "the last time X happened, Y occurred," as if to imply history is destined to repeat itself. Well, "the last time" was only a single event----- it provides no predictive value. It's the equivalent of saying, "the last time I flipped a coin, it was heads, therefore next time it will be heads too." That's obviously a ridiculous statement, but I see people using similar logic all the time.
In the investing world, I can't tell you how often I've heard professionals say something to the effect of "a certain asset class did well during the last recession, so you should own product X to protect against every recession." If they had proof that during every, or even most recessions this particular asset class did well, they'd be on firmer, though not solid, footing. But this is usually not the case. A recent powerful example often overpowers our desire to dig further.
Data snooping refers to when a connection between two or more factors is made after data has been collected. It's given this name because you need to "snoop" through data to find a connection. Although this may not sound like a bad idea, statically speaking, you can generally find a correlation between at least a few data points if you have enough of them.
One of the more famous examples of this is known as the
Super Bowl indicator
. In the 1970s, it was discovered that every year an AFC team won the Super Bowl, U.S. stock markets would have a bad year. Conversely, when an NFC team won, stock markets had a good year. As of last year, this correlation has held up for 40 out of 50 years, or 80% of the time----- a very good rate of prediction. Obviously this is just a coincidence, but if it weren't, I'm sure an enterprising investment firm would develop an investment product (called the Super Bowl Fund?!?) to outperform the market.
Instead, correlations are much more dependable when resulting from a prediction as opposed to being confirmed after the fact.
In the mutual fund industry, companies often tout the performance of their overall fund family's record by indicating something like "60% of their funds are four stars (out of five) or better." Sounds impressive, doesn't it? However, it conveniently doesn't include many of their worst performers, as it's common practice to either collapse or merge them into better performing funds. How do you count something that doesn't exist? If it doesn't serve your purpose, you don't. By burying past poor performance they have in effect recreated the past--to their benefit.
Luckily, there are organizations that track the long-term performance of the investment industry, including investment products in the "graveyard of finance," so to speak. I can tell you with confidence that the results are far worse when you account for survivor bias.
In the review queue
Tim Ferris Podcast with Shep Gordon
Shep Gordon used to be one of the most successful managers in the entertainment business and the person credited with inventing the 'celebrity chef' phenomenon----- a great podcast full of fun stories.
The World's Population in 2100 by Region
by Visual Capitalist:
Although things won't likely work out exactly as the chart predicts, this dynamic chart based on current trends is worth looking at none the less.