When I construct a strategy, it is easy to make subtle dependencies on trends that have existed for a long time.
Sometimes it is legit to explicitly take advantage of the trends. For example, it has been wise to own the index, and shorting the VXX could have returned very favorably. When we do this, we are very clear what we are doing and manages the risk accordingly.
The problem is that the dependency on these trends can be subtle, and, because these trends have been going on for a long time, backtesting will give a false sense of security. But when the underlying trend reverses, the risk can be huge.
For example, in the low realized vol environment in the last few years, leveraging up longing the market or even borrow to short the VXX could have worked beautifully. These show up well on back-testing stats for years and years, but it does not mean future risk is what the stats reflect.
To summarize, the problem is that the dependencies of the strategies on the trends can be subtle and, because these trends have been running for years, just relying on back-testing stats gives a false sense of security, because past performance does not predict future performance.
This gives rise to the next question: what if it is a complex quantitative strategy that does not depend on any well-known trends, but instead depend on esoteric statistical properties of the securities. The strategies work because these stats present a trend, which can be running for many years and gives false sense of security, but can also abruptly reverse and cause huge risk.
But all trends can reverse. I guess the fundamental question then is that how we know which trend, well known or obscure, can catatrophically reverse.