[dlubarov]

Wouldn't that be for fixed-dollar fees? I think here all the costs we're talking about are percentages.

I.e. ignoring taxes, the amount I theoretically expect to exit with should look like

    entry_cost * (return_rate * fee_rate)^T * exit_cost

Where return_rate might look like ~1.1, fee_rate might look like 0.9983 (17 bps), and entry_cost and exit_cost might look like half_spread/price (under some assumptions...).

So I think this comes down to whether T is large enough for that exponentiation to dominate the half-spreads.