The tracker programs calculate expected value only for situations in which you are all-in. Your winnings are counted in other hands.
For example, folding in the big blind decreases your EV by 1BB.
For example, stealing a 5BB pot increases your EV by 5BB/#ofplayers - rake.
For example, going all-in preplop with AA vs XX, with each player contributing 100BB after rake gives you +85BB EV regardless of the actual outcome of the hand.
Therefore, if you lost 4 60-40 all-in coinflips in a row, your EV would be higher than your winnings, indicating a downswing caused by bad luck. If you lose money without having higher expected value, it was probably poor play rather than variance.
This, at least, is my understanding of how HEM (and I presume PT3) calculate expected value in graphs and session statistics.