You may think that good past performance is evidence of future competence. You would be amazed!
Why not pick investment vehicles on the basis of past performance? Agreed, previous good performance may have been accidental. But surely picking a good past performer is at least more likely to be successful than picking at random?
Wrong! Even the regulator agrees, to the extent that all advertising needs to include the statement "past performance is no guide to the future". To quote its report:
"Our analysis of the information available from academic and industry sources and from our own work leads us to believe that both firms and consumers place too much emphasis on past performance information. Past performance is not a useful indicator, for consumers, of future returns. In addition, where past performance is presented selectively - with time periods selected artfully to show past performance in a good light - the information can be positively misleading."
How does it work?
There are many mathematical and presentational tricks that make it possible to build an illusion of performance. Some of them are:
- Selective choice of time period. The regulator tries to neuter this scam by mandating the time period to be used (e.g.five years). But that's OK. Just promote your fund when five years ago was a low point. At other times, keep quiet.
- Selective nurturing (killing bad funds and promoting good ones)
- Inappropriate comparisons (e.g. comparing the performance of a high risk fund, which needs to have high returns, with a low risk fund)
- Data mining (choosing data favourable to your case, ignoring data that isn't)
- The Law of Large Numbers (something exceptional will turn up if you pick from a large enough sample - or, enough monkeys, given enough time, will write Hamlet). Very powerful when combined with data mining.
All the paraphernalia of performance records and tables are valueless unless they are processed by a specialised independent measurement agency such as Morningstar in the US. Whether even that has any value is another, and unresolved, question.