But this argument misses something - what about the number of people being simulated? If there are many more than actual people, the same argument applies. You are likely simulated.
But, you may object, if you simulate a person, don't you need to simulate a universe? Well, yes, but only the bits the person can see. Like in a fps, it doesn't render the entire world from every possible perspective - just one. Much easier (and, dare I say it, more likely than simulating an entire universe). This is just Descartes systematically deceiving demon + how many there are.
A second objection is that it would be more efficient to share the same world across many simulated people. This might be true, but the underlying question remains: to determine how likely you are simulated, we need to consider the simulated vs real people ratio, not the simulated vs real universes.
PS: If we're going all Vishnu's dream, all bets are off. Why should it in anyway be bound by the observations we have gathered? For example, why assume the simulation must be finite - any justification that we can generalise from our own observations is parochial and anthropocentric.
Of course, it is simpler to imagine (and probably, to do), if it's a simulation of the actual universe. So let's use the argument again: of all the simulated universes, how many will be of the true universe, and how many will be bizarre variations (which ours would be)? Of course, we cannot hope to estimate these probabilities for an arbitrarily different parent universe - even less so than to predict what that parent universe would be like! But it seems pointless to simulate an exact copy... for us, iteration-like variations, or of fiction seem more likely.