« Machine translation and global news | Main | Nomograms »
May 23, 2006
The Vanguard Group is a corporation that manages a bunch of mutual funds, and probably does some other stuff too. I have money in several of their mutual funds, and receive performance updates from them. This graphic was in one of those updates:
The idea, I guess, is to be able to see which asset classes have done relatively well or poorly over time: within each year, the best-performing asset class is at the top and the worst is at the bottom. If you pick out a single color, it's barely possible to follow how it has performed (relatively) by seeing where it sits in each year...sort of mentally connecting the boxes into a line graph. For instance, look at the dark blue "LG" (a group of "growth" stocks defined by Citigroup), which was the top-performing set of stocks from 1995-1998 inclusive, but then dropped to third-from-the-bottom for 2000-2002. The last column shows the average gain over the whole time period.
This is a very bad graphic. Although the actual performance for each class in each year is given numerically, the graph itself gives no information about this, which is after all the most important characteristic of an investment. Being at the top of the chart might mean you gained 75% (1993) or 7.8% (1994). An investment that was eighth on the list might be just a bit worse than the one that was fourth on the list (1992) or might be far worse (2000).
The graphic above is horrible, but It is not trivial to make a good graphic that displays the same data. I just spent 30 minutes on it, producing the following:
I tried to use approximately the same colors, but didn't work too hard at it....the color key from the previous graphic sort of works, except that the flesh color on my graph should be hot pink. (I have to get some real work done today, so sue me.) I made the background a light gray to make the light-colored lines show up better; if I wanted to spend more time I would change the light colors to something more intense.
It would take some work to use this graph to figure out "which was the 5th-best-performing asset class in 2001", which is something that is easy to tell from the Vanguard graphic. But at least this [;py conveys some information about the magnitudes of the ups and downs. And look at the T-bills (light blue), just chugging along at their steady 1- to 5-percent gain each year...contrast this with their dramatic ups and downs on the Vanguard graphic.
But if I were Vanguard, I wouldn't send out my version of the graph either. It has lots of problems. A big problem is that the dynamic range of the yearly gains and losses is far wider than the average gains and losses (a fact not conveyed by the Vanguard graphic unless you look at the numbers themselves), so the dots at the end of my graphic, which indicate the average gain, would overlap if I didn't add some offset in the x-direction. Also, I would have preferred to use those dots to provide a legend that matches the color to the asset class, but there's no room since the text would overlap.
I would be interested in seeing what people come up with, as a better alternative to these plots. Why not download the data file and see what you can come up with?
Posted by Phil at May 23, 2006 03:28 PM
Trackback Pings
TrackBack URL for this entry:
http://www.stat.columbia.edu/~cook/movabletype/mt-tb.cgi/459
Comments
I spent a short time in financial sales- getting people into mutual funds and such. We were supposed to use this chart to show people that you can't tell which asset class was going to do best so you have to have some money in all of them (which few people do-- so lots of potential sales!).
In other words, it's supposed to be confusing.
Posted by: Matt at May 23, 2006 04:53 PM
The graphic from Vanguard is most likely derived from the Callan Periodic Table of Investment Returns, produced each year by San Francisco-based Callan Associates. It's quite common in the investment management field because it's typically used to remind investors that no single asset class has produced consistent returns over long-ish spans of time (5-7 years). The jumble of colors and rankings, therefore, serve an additional purpose.
Tufte it's not, and indeed his example of Sparklines from his forthcoming "Beautiful Evidence", in which he overlays several successful diversified mutual funds and finds they track almost identically, may ultimately prove more instructive for investors.
Note also that Vanguard and Callan focus on public markets, not private equity/venture capital or hedge funds, for which the data are much harder to come by and interpret.
Posted by: Claiborne Booker at May 23, 2006 05:42 PM
Recognizing Matt's point, were I actually trying to inform my clients I would graph a cumulative portrait of the asset classes. Doing so shows that the best-performing assets (sv & mv) are neither the ones with the least nor the most volatility. [Of course, I can't post the figure itself, but the formula is just x(t)=x(t-1)*(1+pct(t)/100), where x(0)=100.]
Posted by: mikem at May 24, 2006 09:29 AM
the best chart depends on the message you want to convey. As Matt said, if the goal is to obfuscate, then the original chart manages to bury the signal in some noisy colors.
If the goal is to show the signal (i.e. directional evidence) and hide the noise, then boxplots work wonders. See my post on this topic.
Posted by: Kaiser at May 24, 2006 10:10 AM
If anyone has a figure they would like to post on this thread, email it to me at pnprice*lbl*gov (replace * and * with @ and . ) and I'll add it.
I don't see how boxplots would work here...well, it depends on what you want to show (duh). But if the goal is to be able to visually compare year-to-year performance of specific asset classes, then box plots won't do it.
Posted by: Anonymous at May 24, 2006 01:12 PM