>The chance of this happening due to luck is impossibly low.
The chance of this happening by chance to some investors (though not particular investors) in a sufficiently large group of investors will approach 100%. Also, depending on the density of the relationships between investors and the amount of attention they get for their returns, the odds of one of those investors being within a friend of a friend of yours will also approach 100%.
The existence of runs is not evidence against randomness.
If you flip a coin a hundred times and it comes up heads 98 times, you're probably safe in assuming it's a biased coin.
OTOH, if you flip it several billion times and it has a run of 98 heads, you can't, on that basis alone, conclude that it is a biased coin.
You can't just pull the run out and treat it as if it was the whole universe of data.
The few that exist might be documented (but please enlighten us on where), but if they are, it's most likely because they are so rare.
"I personally invested in a fund 12 years ago which has returned a compound 21% pa return over that period, so I know that they definitely exist."
That's exactly what the GP says - of course there are, but how do you which ones will be, and know you didn't stumble into this one by chance? Your story might just be survivor bias. When I look at the funds my banks offer, there are some that offer 10%+ returns for the last years. Thing is though that they offer dozens of funds, all of which look equally likely in their prospects. How do I know which one to pick? Their yields are normally distributed, just as one would expect.
- One individual or fund doing well in any particular year = random chance
- One individual or fund doing well for a decade = chance is not a possible explanation
Essentially the chance of getting a heads on a coin flip is a 50/50. If you get heads, it's a chance event.
However, if you do 10,000 coin flips and end up with 8,000 heads, either the flipper or the coin have an edge in making heads turn up. The pattern shows that it is no longer a random 50:50 chance.
EMT is probably approximately true, specially with fast algorithmic trading and automated analysis. But the degree of accuracy is key -- some applications are highly sensitive to how good this approximation is, allowing big gains. Also, in practice, there is a "zero sum" part of trading, and that necessarily implies that someone is going to make money and someone will loose (disregarding the other part). Whoever has the best model wins consistently.
"Multi-decade returns of 20% and above"
Do you have any proof of these over the top claims?
I recommend reading http://www.amazon.com/The-New-Market-Wizards-Conversations/d... and the other versions in the series - I think there is 4 now. These are candid interviews with the individuals involved. They are good reading just for the interested entrepreneur.
Some professionals have an edge by developing a model of the market that suits their particular risk levels/cognitive model. As long as that edge continues to work, they continue to outperform by large amounts. The edge might be extraordinary discipline, it might be a allocation size model, it might be a risk-management model. The edge has to exist, and most importantly, it has to match the psychological profile of the individual.
The mistake is not that the EMT is particularly true, it is that amateurs approach the markets with the hope of replicating the results of the professionals, and that amateurs think they have an edge, but they in reality do not.
If the Efficient Markets Theory is to hold true, then in no case should people have multi-year returns well in excess of average. Since this is true, then most people will say that the theory is false. I don't really have a dog in the fight- I don't care either way. But I don't think you can say that it is chance that causes people outperform the market, because there is far too much evidence to the contrary.
As for the 'people studied for years' bit - I don't think that strengthens any argument. The research world is littered with theories that have failed, and length of time spent researching has little correlation with factual correctness. Ultimately, more theories have to fail than succeed in order for knowledge to be gained.