The law of large numbers and either conservative capital allocation or an algorithm that detects higher risk situations or utilization of consistently negative correlated assets should create very predictable and reliable automated system.
so to accomplish this is no easy task the problem all model builders have is they omit the "benchmark" which i learned from construction days is your god on the project. without a "benchmark" how can anyone build a model? it's all random if you don't have a "benchmark" you are forever lost and will never succeed.
secondly compared to the first part the "benchmark" easy would be to find an edge. in the old days we got slippage like you would not believe so add the induced delay of the data feeds the exchange members were giving you and the time delay to call a broker you had a better chance to trade opposite of what your original order was you had a better chance.
that made me a reversion to the mean dropping knife catcher type trader, it taught me to fade everything and when it didn't work out flip it quick. luckily now days you can as nimble as you want but you have to have everything in line to make it all work to the efficient level that is available to all traders now not just members.