My personal list of strategy characteristics required to start trading, in descending order of importance:
1) An explainable, cogent reason why the strategy’s signals are profitable (in bulk).
2) Consistent performance of a random subset of a specific instrument type over time (out of sample) and across random selections of instruments within the asset type.
3) Drawdowns as a percent of account less than 10%.
4) Limited to no human intervention required.
5) Trade sizes that can scale.
6) A high enough frequency of trades such that skepticism in the strategy’s operation is not called into question.
7) A uniqueness or eccentricity in the strategy’s technique such that the likelihood of discovery or loss of efficacy is limited.
So far I have built strats that satisfy 1, 4,5,6 and 7. It’s 2 and 3 that are the problem spots.
Some of my rational:
1) If you don’t understand it – you won’t trust it. Blackbox style NN algos that are completely opaque as to logic – how are you going to trust them when they start a losing streak?
2) Sector based or venue based selection is good. Grains don’t trade like energy futures. Utilities don’t trade like tech stocks. But you should be able to randomly select from within your sub sector or category and perform adequately with any sub selection. And of course, a strat should be able to test in and out of sample, walk forward style, over vast amounts of data.
3) After you lose 10% of your bank, you’re gonna question the strategy’s potency.
4) I don’t want to sit in front of this computer screen(s) and trade. Not my thing.
5) FX you can trade for pennies, or millions of dollars. mini-futures, a few highly liquid securities, an option or 10, all scale very well. Bond trading? Low liquid stocks? Full contract futures? Not so much. But you want a strat that you can pile into after you start winning. So, the instruments needs to scale.
6) A strategy that trades 5-50 instruments, at least once a day (in total), means it’s working, and you can be fairly comfortable in its operation. Once a week? Sure for investing – not trading. More than 100 time a day? I doubt anyone could walk away from a beast like that and golf the mornings away.
7) Belief in your strat’s ability to profit is paramount. A generic block of logic you pulled out of S&C magazine is not a candidate for trust. Something in your strat, its logic, instrument, time of day, scaling, stops, basket technique, pairs algo must be unique ( as far as you know) so that you can trust it when it starts to lose.