Well, I will give you the biggest upset that I have ever seen. Once in club chess rated at a long time control GM Darryl Johansen lost to a 1300 rated player. I don't know what Johansen's rating was at that time [it was late 80's early 90's from memory]. That is ONCE out of 1000's of rated club games.
So I will say 0% chance in the sense of it being so asymptotically unlikely.
I think a 1300 player would struggle to beat Gelfand after that blunder to be honest. That's even if they found the Qf2 move that the strong GM commentators didn't. How sure are you really of beating Gelfand from here?
Yeah, the point is that they don't blunder from easy positions. Blunders come from positions where there are complcations, and normally when you are poor positionally.
A 1300 even if they can avoid tactics, isn't good enough positionally to put a GM under any sort of pressure.
It could possibly happen in a simul where the 2700 player was playing maybe 20 players and somehow hung his queen or something.
Actually it is 0%.
Not to say that it's impossible, but to express the chance you need a much higher accuracy/more decimals.
From my experience of watching chess players each gap of 200 pts is 1 difference in chess technique that is massive. For instance one thing that I notice is that 2000 players are much, much worse than 2200 players at dealing with counterplay in a won game with tactical complications; the 2200 player will win such positions easily and without fuss. So 2700 - 1300 = 7 of those "technique blocks". That is a heck of a lot.
If such an upset does happen it is usually a quick game with a serious opening inaccuracy through carelessness. The game was pretty quick in the case of Johansen losing - about 20 moves. Of course Johansen is not a 2700 GM. So the task is harder still.
A player who actually has a SUSTAINED rating of 1300 has 0% chance.
I beat or draw 1300's 100% of the time in otb play and have a rating of about 1700 although I play much better than that. A 2200 player beats me 15-1. A 2700 player beats a 2200 30-1. So what does that tell you?
Or the 2700 is Chucky in the mood to "chuck" pieces your way for free 
In OTB games, I've often seen 2200 players win back against 1600-1700 players after blundering a piece.
So 0% for me too 
Why not simply let someone call the GM on the mobile during the game? 
But wins like this are too unlikely to happen in tournament practise. There are simply way too few games between players of that rating difference.
In a big simul, such an upset is of course possible. (There are underrated guys at 1300 that actually can play a little chess and if you simplify quickly and then drop a piece being a Super-GM is not much help)
Well I don't know about a 1300 beating a 2700 but in my case a 1750 player I have played some 1000 rated players who play so loose that the unexpected can happen (Akeetam for one). Lost to a 1300 player playing too safe recently and been so up in material in one game my position became swamped by my own material.
Why not simply let someone call the GM on the mobile during the game?
But wins like this are too unlikely to happen in tournament practise. There are simply way too few games between players of that rating difference.
In a big simul, such an upset is of course possible. (There are underrated guys at 1300 that actually can play a little chess and if you simplify quickly and then drop a piece being a Super-GM is not much help)
Even in a simul, a win is rather unlikely, unless the 1300 guy is seriously under-rated.
A draw might be possible (GMs are happy to draw in simuls, as it doesn't affect their payment and they can go home earlier )
try a large stick lol
My answer to this question ties in with the fact that those online players who strategically play others 400 or so points lower than them are not necessarily going to make a profit in points in the long run, because even though they could get 25 wins in a row, all of that will eventually be returned by an "unexpected" loss.
I don't really think anything is impossible -- maybe a grandmaster could pull a stripunsky, and hang his bishop and rook. True, even then, the 1300 might still not win, but if we had eight thousand billion tries, maybe one time it would happen.
You know they say that if a monkey was forced to type for an infinite amount of time, eventually the random strokes it makes will "just happen" to be identical to Shakespeare's Hamlet.
Well I don't know about a 1300 beating a 2700 but in my case a 1750 player I have played some 1000 rated players who play so loose that the unexpected can happen (Akeetam for one). Lost to a 1300 player playing too safe recently and been so up in material in one game my position became swamped by my own material.
You talking about online ratings? Because I wouldn't expect to lose 1 game in a hundred (accurately) rated 1000 in online turn based on chess.com.
I'd be impressed to see a 1300 player beat one of the strong computer programs in my above diagram, and that's a bad blunder for a GM, which wasn't easy to capitalise on.
Sure the 2700 player could touch the wrong piece accidently and lose a queen. But unless the 2700 player makes a horrific blunder, forget it.
Scottrf: Hell, I thought Anand should have been forced to play out the position!
Yes there was but a higher rating your opponent the lower the chance you can win on it. This is just a logical one.
Of course, I'm talking about games when both players want to win.
I also don't buy %0 percent chance, because I think I can get a win against a Boris Gelfand type blunder.