Frank Sortino, PhD, occasionally tells the following story:
There was a pension plan that was looking for a Japanese market investor. They found one who had superior performance. Unfortunately, he also had a high tracking error. When their concerns were presented, the manager explained that much of his performance came from the fact that he avoided the banking sector; likewise, the high tracking error could be attributed to this decision. While this seemed to make some sense, the pension fund trustees were still concerned about the tracking error, to which the manager replied, "okay, how many bad banks would you like me to invest in?"
Question: should the benchmark have been void (i.e., ex) the banking sector?
[pause, while you ponder this question]
Answer: no! The manager's strategy includes the entire market; his tactical decision is to avoid the sector. The benchmark should not be adjusted for tactics.
Question: why not? Why shouldn't the benchmark be ex banking?
Answer: if it was, then it would be difficult to determine if this was a good decision on the manager's part. Perhaps banks soared during the period; by having the benchmark ex banks, this wouldn't be known. If, however, this was a good tactical move, then the excess return (as well as, presumably, the attribution analysis) would reflect this.
Question: what if the CLIENT requested that the manager avoid a particular sector; would the benchmark then be void the sector?
Answer: yes! Because now, the strategy excludes the sector, and the benchmark should reflect the strategy.
Question: other than buying bad Japanese banking stocks, might there have been another solution to the prospect's concerns with the high tracking error?
Answer: yes! Why not run the tracking error against the benchmark without banks; granted, the benchmark for return comparison (and, for all other statistics, including tracking error) should be with banking; but for the purpose of determining if this tactical decision was the source of the high tracking error, such an analysis would be helpful. Going forward, the manager could report both versions of tracking error to the client.
Related example: A GIPS(R) verification client of ours' composite description reflects exposure to stocks, bonds, and commodities. The benchmark should include the strategic allocation for each asset class. However, when we look at a representative portfolio, we discover real estate (through REITS), too. Should the benchmark include a representation from this asset class?
Answer: only if real estate is, in fact, part of the strategy. If, however, it's a tactical move, to take advantage of a short term opportunity, then we would not expect to see real estate present. But, if we find that going forward, real estate has become continually present, we would argue that it now seems to be part of the strategy (i.e., a broadening of the strategy), and would now need to be included.
Make sense? Please chime in!