The credit for today's headline goes to Andrew Gelman, who said something like that when I presented the following chart at his Statistical Graphics class yesterday:
With this chart (which appeared in a large ad in the NY Times), Fidelity Investment wants to tell potential customers to move money into the consumer staples category because of "greater return" and "lower risk". You just might wonder what a "consumer staple" is. Toothbrushes, you see.
There are too many issues with the chart to fit into one blog post. My biggest problem concerns the visual trickery used to illustrate "greater" and "lower". The designer wants to focus readers on the two orange brushes: return for consumer staples is higher, and risk is lower, you see.
The "greater" (i.e. right-facing) toothbrush is associated with longer brushes and higher elevation; the "lower" (left-facing) toothbrush, with shorter brushes and lower elevation.
But looking carefully at the scales reveals that the return ranges from 6% to 14% and the risk ranges from 10% to 25%. So larger numbers are depicted by shorter brushes and lower elevation, exactly the opposite of one's expectation. The orange brushes happen to represent the same value of 14.3% but the one on the right is at least four times as large as the one on the left. As the dentist says, time to rinse out!
The vertical axis represents ranking of the investment categories in terms of decreasing return and/or risk so on both toothbrushes, the axis should run from 1 to 10.
How would the dentist fix this?
The first step is to visit the Q corner of the Trifecta Checkup. The purpose of this chart is for investors to realize that (using the chosen metrics) consumer durables have the best combination of risk and return. In finance, risk is measured as the volatility of return. So, in effect, all the investors care about is the probability of getting a certain level of return.
The trouble with any chart that shows both risk and return is that readers have no way of going from the pair of numbers to the probability of getting a certain level of return.
The fix is to plot the probability of returns directly.
In the above sketch, I just assumed a normal probability model, which is incorrect; but it is not hard to substitute this with an empirial distribution, if you obtain the raw data.
Unlike the original chart, it does not appear that consumer staples is a clearcut winner.