When one drives a car one obviously looks ahead out the windshield as well as the mirrors and through the back window to what is behind us, with priority given to the former. Historical testing is all well and good but not where gains or losses are made.
If trading was like driving a car, then everybody would be able to do it. A driver knows with almost complete certainty what is coming next. Car physics are simple and even intuitive, and can be described well by a few simple equations. Trading is the polar opposite of this, with most outcomes being close to unpredictable. It is a chaotic, dynamic, competitive game. In this kind of environment, empirical modeling ("Historical Testing") is one of the most powerful tools for decision making. It's more than just "well and good", it's foundational for quant trading.
First everyone cannot drive a car. Anyone who thinks that hasn't driven America's (and all other nations) roads. Second of all, trading as you've pointed out is chaotic, dynamic and competitive, which makes what happened historically to have very little bearing on what will happen in the future. All testing does is tell you don't trade this if historically it loses. If it shows a strat to be a winner only forward testing, with real money on the line, will say yes or no - no matter how many years/trades/out of sample/etc etc.
Hey, you said it yourself... That means looking in the rear-view mirror has very little value compared to what is ahead of you.
LOL. You know what I mean tho This is statement is wrong at a very deep level. Everything that is predictable to any extent is by definition a function of what happened in the past. In other words, from the point of view of someone living in the present, what happened historically is the only thing that has any bearing on what will happen in the future. To the extent that trading is not gambling, those trade decisions will be based on past data. It is the nature of living in a universe where time flows forwards. It's not black and white like that. You need to understand the concept of statistical significance to properly use backtesting. It's like science. No strategy can ever be "proven" to work in the same way that 1+1=2 is proven, but you can get useful results.
Only if the testing is unable to exactly consider each and every detail of live trading. For example, testing may not be able to approximate the integrity and honesty of a broker. If the data is clean, if the spread can be approximated, if slippage can be approximated, then testing can be golden if careful not to overfit to curve (reserve some data for pseudo-forward test). Forward live tests can help approximate slippage and spread conditions by comparing to estimates. Those are most critical the shorter the time frame, the smaller the average win. Longer term strats (larger average stop loss) not as much reliant on those two factors.
If the concept of fractals is valid, you should be able to backtest (if the test has integrity as described above) on shorter time frames, being aware of how times of day might affect your outcomes. You would make a compensation on slippage and spread (or commission) to mimic your longer term conditions . Better yet, just test the shorter times when your weekly strat is supposed to be in the market (making money). In theory, if your main strat works, it should provide a tail-wind for your similar sub-strats. After that, some software to help. Excel macros for example. Or any platform that can speed up pseudo-manual trading on historical data.