There seems to be a lot of interest in discussing multi-period performance attribution. In one of the LinkedIn groups, we’ve had quite a dialogue going on for the past few weeks, spurred by questions raised by our colleague, Kathleen Seagle. I thought I’d briefly comment on the topic here, and introduce a research effort we’ve begun.
Several years ago, The Journal of Performance Measurement(r) began to publish articles that provide methods to “link” arithmetic results across time. Arithmetic attribution is, to use today’s lingo, “linking challenged.” That is, residuals will result, and must be “smoothed out.” Authors such as Jose Menchero, David Cariño, Andrew Frongello, and Damien Laker provided details on their methods.
But for how long a period should we calculate multi-period performance attribution, and should these results be annualized?
Geometric Attribution Myths
While geometric attribution isn’t “linking challenged,” it is “single period challenged,” because residuals result for any period in which it’s used. As a result, smoothing methods are required. Jose Menchero’s Journal article on geometric attribution detailed this fact, and he provided his method to smooth the results out. And so, the myth that geometric doesn’t have residuals is just that … a myth (just like the myth that geometric models don’t have interaction effects).
Multi-period Performance Attribution Mini-Survey
TSG decided to launch a “mini survey” to explore a bit about the subject of multi-period performance attribution. There are only four questions, plus a fifth for you to offer whatever additional comments you would like. We’ll take answers through the month of August (i.e., it ends August 31). Please join in! We’ll publish the results in early September.
On this topic, I will shortly begin to do some empirical research, in an effort to get some “hard facts” that may provide further assistance. Stay tuned!