Yes. The objective can drive stddev to small value to inflate the Sharpe ratio.
So far, in my experiment this doesn’t seem to be an major issue yet.
My issue is that it seems to optimize for high turnover portfolio, this is even after I subtracted transaction cost from the return in objective function. I am trying artificially larger transaction cost to see if it helps.
There are papers suggesting machine learning based stock prediction typically will optimize toward high turnover, and many so called high return will be gone after taking into account the transaction cost.