Context is everything.

There is a HUGE difference between:

(A) short term random fluctuations...
In the context of a strong correlation...
To a meaningful historical level.

VERSUS

(b) Purely random independent events like coin flipping.

People like YOU do not understand the difference...
And constantly MUDDY the WATERS...
To the benefit of Quants...
Who exploit this subtlety to make money...
A LOT of money over time.

Octoputz, please keep posting as much as possible...
You are great for my Bottom Line.

^Please never post in this forum again.

True, but regression to the mean applies to both.

+1

grunch from like, first 5 posts:

I was thinking about this in the shower yesterday for some reason and I was thinking that this might have something to do with it... that a coin flipped for x times will average half tails half heads is working with the assumption that at flip1 there are 0 heads and 0 tails. But if we start at f1 with 0h and 7t, this would not be the same thing. In this scenario, I think we would be wrong in thinking we should average half tails and half heads, instead, we should expect half heads / half tails+7 (after 100 flips, we would expect, on average, 50h and 50t + 7t). So the more flips the less this 7t start would make a difference (ie after 1,000 flips we would get on average 500h and 500+7t, vs after 2 flips 1h and 1+7t).

You are both right so far but your rebuttal to the teacher's is missing one thing. The teacher's comment of regression to the mean is a very good one. Your rebuttal about independent coins is necessary but will leave readers not fully convinced why then does regression to the mean happen? The key assumption of regression to the mean is the infinite tries.

Regression to mean does happen and is exactly the mean at infinity tries. The thing is that the distribution of the outcomes are completely random and you can only expect the "remainder" outcomes if there is a finite amount of tries. Since it is an infinite amount of tries, you can Never(infinite) expect to have more heads if you've been getting tails.

Adrianrain - 9 hours ago

* Rank: 1 / 5 (1)

I use a trick all the time and it works for me. If i need to flip a coin and i want it to be heads for example. I will keep flipping it till i get two tails in a row then I will go out to where I have to go and flip it again. Three tails in a row is hard to get but two tails followed by a heads seems easier.



This is from an article about how with practice one can achieve a more favorable outcome of coin flips. Also once had a friend tell me that he checks recent hand histories to see how often an ace hits the flop to predict future outcomes. Que facepalm.jpg

For some reason this concept is more obvious and intuive after a hot streak than after a cold one

I like Muddy Waters, good band imo.

I use a trick all the time and it works for me. If i need to flip a coin and i want it to be heads for example. I will keep flipping it till i get two tails in a row then I will go out to where I have to go and flip it again. Three tails in a row is hard to get but two tails followed by a heads seems easier.

I keep reading things like this and I think people keep having slightly the wrong idea about it (imo).

If you flip a coing 5 times and it comes out heads every time:

- The chance of heads flipping next is 50%
- The chance of tails flipping next is 50%

However the total heads/tails flips will not even out over the long run. Over the long run you can expect to have +5 on the heads count still!

The only thing that has changed is the resolution of your graph, where so many flips have occured that 5 intial flips, or any initial flips for that matter become drowned in the graph as that sample taken from the entire set of data no longer has the significance as it did when it represented 100% of your data set.