DesertCat is *vastly* oversimplifying the case of backtesting and imo misrepresented it b/c it can be a very useful tool ****IF USED CORRECTLY***
Quote:
Originally Posted by DesertCat
I'm saying backtesting proves nothing that noticing that 7s are hot does, i.e. that random patterns don't persist no matter how carefully you match an algorithm to random historical events.
that is an example of BAD backtesting (finding patterns that worked and presume that they'll work going forward). in other words, it's datasnooping and is a known bias.
"proper" backtests would be, for example, thinking about fundamental relationships regarding the economics of interest rates, economics, and market outcomes and then creating a logical rule based on those relationships and their hypothesized market outcomes. then, testing that rule to see if it worked.
then, thinking about similar rules that may also have an ability to explain the variance in market returns and constructing a signal based off of those rules to test to see if it works.
the key is starting with logic and ending with the backtest. backtesting to FIND returns is exactly as DC has let on; however, thinking through fundamental relationships (i.e. ones that are likely to last over time or at least remain important/significant) and creating trading rules based off of those and then testing them for efficacy is a method of generating successful systems.
i know i keep coming back to this but just see bridgewater associates or ray dalio. that is literally all they do (and they do it basically exactly as i described above).