PAST PERFORMANCE DOES NOT PREDICT FUTURE RESULTS. That line is on every piece of investment literature that goes out to clients, they get it, but often times advisors don't. Past performance of any investment or methodology is virtually meaningless, it tells you how something performed in the past, but not why it had that performance, which is the critical component in figuring out how something is likely to perform in the future.
Long Term Capital Management is a classic example. A few years of great performance followed by kaboom. Their performance came from a more sophisticated knowledge of arbitrage opportunities and more sophisticated systems. It was only a matter of time before the rest of Wall Street caught up and those opportunities disappeared.
Asset allocation is another example. From 1982-99 it did great, that didn't tell investors anything useful going into 2000-2002. From 2003-2007 it also did great, that didn't help investors in 2008. The real question is why did asset allocation do great? It did great because the market was going up, since the market doesn't always go up, and since nobody can predict when it is going to go down, the past performance of asset allocation is useless.
The same issue can be found in backtesting tactical methodologies and systems. We have a very successful counter trend system for Treasury Bonds. However, adding an inverse side (betting Treasuries are going to go down) historically reduces returns and increases risk. If I relied purely on the backtested results I would be tempted to remove the inverse side and have the system be long only. We decided to keep the inverse side in because we asked why did the returns work out the way they did in the backtest. The reason is because over the entire duration of the backtest Treasury Bonds were in a bull market. However, even in this environment, the inverse side was still slightly profitable. Will Treasury Bonds continue in a bull market? Probably not. Therefore, having an inverse side to the system would have hurt performance in the past but will probably help performance in the future.