Despair wrote:

@Jebus: I actually think you are wrong. Seeing the performance on all data before you are finished with the algo will give you an impression of what might work and what not and you will (even if it is subconsciously) adapt the strategy accordingly.

But if you have 50% of the data and u make system A that looks decent enough to work with.
Ur gonna start optimizing that on the 50% of the data and then when ur done u check on 100% to see whats gonna happen.

That same System A, whether its made on 10%, 20% 80% or 100% of the data, if u make it the same way, with the same variables, it wont matter at all, because its the same exact code?

Let me also go through my process of making strategies

Part 1: “The trade”
I cherry-pick a fantastic piece of data (I choose a wonderful perfect trade setup in the chart) and i create my strategy based on that single trade. (this is like >1% of the data)
If i get a good entry/exit and it dosnt look like shit just before/just after that trade, i check the 100% of the data. If it all looks good on the 100% i go into part 2 (i dont wanna spend hours optimizing and fixing my strategy if it looks like shit in 100%.. i dont wanna polish a turd, if u get what im saying..)

Part 2: “The optimization”
I pick 30-50% of the data (usually the first 30-50% of the data provided) and i optimize that system, everything thats going into the code is already in there, i just optimize the variables from say RSI 14 to RSI 12..

Part 3: “The flippening”
I call this the flippening cus its when i flip my optimized system, onto 100% of the data to see what happens.

I guess ur right that im gonna have a bias that says “This might work”, but i wouldve had that bias if i had only the 50% data to check.. i would never optimize an entire system based on data worth 1 trade. Does this make sense?

@Grahal MAYBE you could take the system and start tweaking and forming it into something that might work.. but in my eyes i have already curvefitted that system, and its not robust at all. To make it more robust i would have to “fit the model” again, but it is already fitted for 50% and turned out as not robust at all..

This to me is a big red flag and optimizing this even more would seem like a very dangerous thing to do!

“Anyone can make money in the markets, but only the very best is able to keep that money” – Someone

Meaning i would rather be safe then sorry! And curvefitting in this game is suicide! Better to stay away is my philosophy

Edit 2: @Despair when i optimize the strategy i only optimize with variables within a short range from the original values. (If i used RSI 14 for my code i wont ask for RSI between 2-100, i would ask for RSI between 11-16 just to check if its much better or much worse when i change some values up/down.