Keep it simple

Keep the Algorithm simple. For your system to hold good for future, the core of your algorithm needs to be founded on common sense principles. Throwing few indicators and using optimizers to come up with some model may not help too far into future. Moreover, it makes your logical debugging of results a nightmare. Also, more indicators/components does not automatically improve the predictive power of your Algo, they may cancel each other out too.

Remember, our objective is not to solve the market and come up with a master key which work in all conditions and all securities. Goal is to design a model which earns decent enough return on investment (compared to the opportunity cost), given the consistent application for long enough duration for a given security.

In my opinion, it is better to have handful of Algos targeting different market cycles (complementing each other) rather than trying to get to an all-weather model. Every Algo will have its favorable and adverse market conditions, understand them and accept them. Those adverse conditions are part of your Algo package in the other words treat it as normal loss.

It would be tempting to keep adjusting the model once we start seeing it in action every day. Make sure you start tweaking the model only if the results do not align with favorable market conditions as expected.

Nothing teaches the game like experience. Come up with simple model which you believe should work in normal market conditions, back test it for reasonable period of time, use a fraction of your capital and start the journey.