When a trading strategy is presented, whether it be in a meeting room of a top trading firm, an online video selling a method, or even in one's personal research process, a backtest is commonly given as a reason to bolster belief that a strategy will work. A backtest by its construction looks to the past when the question is really about the future, therefore it is not sufficient to answer the question on its own.

The validation of a strategy relies on a process akin to the scientific method. In this analogy, a backtest constitutes one experiment, but that alone would not be sufficient; it needs more. More would include (but is not limited to):

Design your experiment, including a pass/fail condition before data is examined. If you have multiple variants you want to test you must account for that in experimental design, lest you fall victim to p-hacking.
Identify and stress the mechanism: does your strategy work with point-in-time data, realistic transaction costs, realistic trading volume? Is it susceptible to regime change?
Where does this idea come from: is this a published anomaly, where public knowledge will eat away the edge?
Breaking up your data into two or more samples: one for training, the rest for testing and validation.
Using the real data to synthesize similar data and testing the strategy on that. Synthesis is more likely to filter for big anomalies that could be the wins that flip a loser to a winner.

A strategy that survives all scrutiny may have some merit. At Caliper Trading, we do not sell strategies, we sell scrutiny. Happy Scrutinizing!

This is not financial advice.