On Returns-Based Style Analysis

Sometime in the next few weeks, I am going to dig into my pre-2003 [pre-RealMoney] files and see if there is anything there to share with readers.  Most of my best stories I have already told in my various series.  The one I will tell tonight I don’t think I have told.

In 1994, we had a problem at Provident Mutual’s Pension Division.  Our main external equity manager was having a very lousy year as value managers that focused on absolute yield were getting taken to the cleaners.  This was after a few years of poor performance — the joke was, given the great performance of the past, “Hey, can you develop the 19-year track record?”  (The last 5 years as a group were horrid, but the previous 14 were great.)

Aside: there aren’t many absolute yield managers in equities today.  Back when dividend yields were higher, and corporate bond yields were higher, both absolute and relative yield managers flourished as interest rates and dividend yields crested in the early 1980s, and the stocks paying high dividends got bid up as interest rates fell, much as the same thing happened to zero coupon and other noncallable long duration bonds.

The process started with a call from a manager of managers who proposed that we start up “multiple manger funds,” where we would be the manager of managers.

This offered several advantages:

  • It offered us an easy out with our long-held failing manager, because we are not firing them, just making them a portion of the assets in the value fund.
  • It would make eliminating them easier in a second step, with less PR damage.
  • It would make us look like we were taking action and control in a new way for our clients. (They loved it.)
  • As it was, we did a good job selecting managers, and the funds performed well.
  • We could negotiate lower fees with the managers,
  • It gave us a great marketing story.
  • Our margins and growth improved.

I was critical to the process, being the only member of the team with investment expertise.  Everyone else was a marketer or the divisional head.  (I take that back, one member of the marketing area was genuinely sharp with investments.)  After we chose the managers, I set the allocations.

Now onto tonight’s topic (what a long intro): At the beginning of our relationship with the manager of managers, they did a traditional holdings-based analysis of how a manager managed assets.  About one year into the process, they introduced returns-based style analysis.

Though the Wikipedia article just cited has a bevy of errors, it will still give you a flavor for what it is.  Let me give my own explanation:

It takes a lot of effort and wisdom to look at quarterly portfolio snapshots and analyze what a manager is doing.  You almost have to be as wise as the manager himself to analyze it, but many fund analysts developed the skill.

But returns-based style analysis offered the holy grail: we can understand what the manager is doing simply by comparing the returns of the manager versus returns on  variety of asset indexes, using constrained multiple regression.

The idea was this: the returns of a manager are equal to his alpha versus a composite index that best fits his performance.  Since we were dealing with long-only managers, the weights on the index components could not be negative.

The practical upshot to the manager of mangers was: “Whoopee!  We can analyze every manager under the sun just by looking at their return patterns.  No more time-consuming work.”

After the first meeting with the manager of managers, I expressed my doubts, and asked for a special meeting with their quants.  A week later, I had a meeting with a few members of their staff, of which one was the quant, a nice lady 10 years my junior, who I felt sorry for.  She started her presentation at a very basic level, and asked “Do you have any questions?”  I asked, “Isn’t this just an quadratic optimization problem where you are choosing weights on the convex hull?”  She paused, and said, “Oh, so you *do* understand this.”  The meeting ended son after that — we agreed on the math, and in math, there is no magic.

But that placed me on the warpath; I genuinely felt the advice we were getting had declined in value.  I wrote a 16-page report to our manager explaining why returns-based style analysis was inferior.

  • There is no way to correctly estimate error bounds, because of nonlinear constraints.  (Note: two years later, I guy came up with an approximate way to do it in an article in the Financial Analysts Journal.  I called him, and we had a great talk.  That said, approximate is approximate, and I haven’t seen any adopt it.)
  • Because many of the indexes are highly correlated with each other, small differences in manager returns make a huge difference in the weight calculated for each index.
  • If a manager is changing investments because he senses a factor like market cap size or valuation is cheap, it will get interpreted as a change in his index, and will not come out as alpha, but as beta.
  • If I don’t believe that the CAPM and MPT are valid, why should I believe this monstrosity?
  • And more… I hope I find my 16-page paper in my files.

After six more months we terminated the manager of managers, and hired a better one.

  • Lower fees
  • Lower fees from managers (they had greater bargaining power)
  • We reduced our fees to clients
  • Better marketing name
  • Holdings based manager analysis

After that, things were much better, and we continued to grow.

My years at Provident Mutual were exceedingly fruitful — this was just one of many areas where my efforts paid off well.

All that said, there is no way to fix returns-based style analysis.  It is a bogus concept and needs to be abandoned.  Those who use it do not grasp the limits of econometrics, and are Sorcerer’s apprentices.

PS — Need I mention that the originator of the idea, Bill Sharpe, is not all that sharp with econometrics?  He’s a bright guy, but it is not his strong suit.

PPS — there are not many actuaries with a background in econometrics.  That is why I have written this.


1 Trackback