Statistical arbitrage is not true arbitrage because it does not deliver a guaranteed profit — in fact many statistical arbitrageurs have made large losses. The term is often shortened to statarb. Although statistical arbitrage is often described as being synonymous with pair trading, it is probably more accurate to say that the terms overlap. Only probably, because it is not possible to lay down an absolute rule when usage varies as much as it does.
Statarb strategies are market neutral because long positions are matched against similar short positions. This does not mean that they are low risk, because they are, in effect, highly geared because to the low cost of the combined trades.
Statistical arbitrage is also sometimes described as a mean-reversion strategy, because of its reliance on the return (at least often enough to make a profit) to historical patterns.
There is a certain similarity to chartism, but statistical arbitrage is more sophisticated (it uses the methods of econometrics) and more credible. What they do share is a reliance on the existence of violations of weak form market efficiency.
Most violations of market efficiency, especially weak form efficiency, are likely to be small and transient. It is therefore unsurprising that statarb strategies tend to be short term and involve taking positions that are very sensitive to the movement in price of individual securities.
Any failure of efficient markets is interesting from the point of view of the theory of financial economics. A no statistical arbitrage assumption can be useful for deriving financial theory (just as the no arbitrage assumption is), and the techniques of statistical arbitrage are useful for empirical tests of weak form market efficiency.