In theory, the regression is always happening. The problem is that it happens so slowly, and with such high variance, that it is difficult for us to see in the short-term. Here’s how most gamblers think of it: If you have a coin that lands on heads 10 times in a row, how do you know that it isn’t just regressing from 10 tails that happened before? How do you know that the coin isn’t at it’s mean after those 10 heads? Maybe it just “evened out” and now you’re betting on a completely random coin. You don’t know.
In reality, the fact that a coin lands on heads 10 times in a row doesn’t mean that the coin is “uneven” at all. It doesn’t mean anything. As Guynoire said, you would expect it to be 10 heads ahead for the rest of its life. If you flip that coin 100,000 more times it will average 50,010 heads and 50,000 tails. But, as the number of flips increases, those ten flip become less significant. The coin still may exhibit a bias, but the percentages become smaller as the number of flips increases. And don’t forget, the percentages can approach the expected 50/50 results even though the difference between the number of heads/tails is increasing.
This theory works in reverse as well. Imagine that the coin has been flipped 100,000 times before you flipped 10 heads in a row. You would expect 50,000 heads in the past, so now it has 50,010 heads and 50,000 tails. The coin is still at 50/50 and there is no reason to expect it to behave any differently than before.
-Sonny-