Howard,
Your thinking is basically the way that I have done it on my MOC style system, particularly in the past. I called it “Walk-Backward” instead of walk-forward testing.
I approached it with two schools of thought:
1. That more recent market conditions are just not as they were ten years ago or more. I mean in terms of the volatility, speed at which it changes direction etc. Algo, PC, Robinhooders, who cares for the reasons why, but it is fair and reasonable to say it likely works faster or differently in more recent years. Therefore I liked to backtest or optimise on more recent periods of time. Say the last 5 or maybe 10 years. Then I will extend it out to 15 or 20 years after that and see how it goes in the backtest. Does it hold up over the longer term as per how it was designed for the more recent years?
However, I also think that this method/thinking may not apply too much because….
2. I have more recently been of the school of thinking that the age-old methodologies and what makes the market work are still as solid today as they were decades ago. I mean price action in the chart, traditional charting patterns, candlesticks etc and how they translate from what you have seen recently, to what the price action is likely to do tomorrow. So in this thought process, this is what tells me walk-forward or walk-backward testing is a good thing, but it just can’t beat finding a system/method that works well through the entire period with very little optimisation by leveraging the age-old methodologies.
And on this second point is exactly where I have landed with how my MOC system is built now. In the past, I would be facing the problem you mention. Work well in say, 2000 to 2010 or 2015. But then start to lose steam in 2015 to 2020. I would blame it on “reversion not working much any more” or trading apps/algos and perhaps that my parameters needed to be “tuned” for the more recent conditions and not to panic so much if not tuned for the conditions of ten years ago. Generally I found that I was probably using too many ‘indicators’ stacked on top of each other to meet entry conditions to make it work. Too many whacky analysis methodologies (which absolutely have their place) but you get lost in thinking they are going to be the holy grail in opening you up to a new way of thinking and finding a new edge for modern times. This almost always led me to failure.
Only in my rediscovery/redesign over the last few months have I gone old-school, back to the basics about price action. No fancy indicators, just moving averages or the basics of how price action works in recent past to what it is likely to do tomorrow, and then bomb it with a massive number of trades, high leverage and smaller exposure per position in my overall portfolio.
You can see the results in my short system stats that I have posted here. You can see that in general, over the last 15 years or so the concept has generally worked in the right direction. Yes, sure, I agree there are the big peaks in 08 and 20 however in the flatter periods it is still moving in the right general direction. Also my two Long systems complement it well to help smooth out the ride.
So I would say that I know where you are coming from and can see the logic in doing walk-backward testing, however I am more of the opinion these days that if a method is only working well in more recent periods that perhaps it may be prone to failing into the future as well, and that perhaps you have not discovered or using a very stable/dependable/robust method within your system.
