@aIgorithms Yeah look like that.

@M19HAU Maybe you found more efficient way to overfit your data :)

@MichaelENewton1 Yeah I tried to keep it simple for reasoning :) (otherwise I would have tendency to add some fat tails distribution, autocorrelation etc. etc., not ideal)

@Roca122 Yes for sure. That's the way, not much other things you can do. (In blackbox approach, maybe some trade distribution, skew penalty etc.) But main points was more simple, one time ranking people see on internet. Is quite biased :)

I don't know/remeber too much detail about their structure. I am not a fan of total blackbox approach. (Not saying it can't work, but I think it's a disadvantage) Incentive.. Yeah good points. Maybe some geographic arbitrage and no barriers may help them hire people. But is it enought..

@LhajjZehwani Don't bother with that. Just buy some Fartcoin, sell some "free profit" options or some "ICT secret patterns" will do the work for you :)

Yeah autocorrelation, skew etc. are one issue here. Second issue is that to have true p-value. You have to know how much traders (comparisions, backtests etc.) was considered. And then use some p-value correction technique. One of best knows is "Bonferroni":

@Roca122 If we select at lest positive results from previous part (only 3 traders continue here): - 1 edge trader is here - 2 noise traders are here as well - edge trader has bad luck on this sample, sadly..

@PtrPomorski @Alice_comfy Not sure what's their goal though. If it's uncorrelated market neutral. Then not beating SPY on total return or even on risk adjusted basis may still be fine. But yeah, really don't know how it went outside of this sample...

One can even calculate e.g. confidence intervals given some assumptions and use that. (It's not so easy to get those assumptions right, given nature of returns, multiple comparison adjustment is needed etc.) But point remains - a lot of randomness. Even more in practice (gaming of metrics etc.)

@dSystems_ai Obviously less noise traders in 2 years sample.

You can do things generally good for you: - without reading / believing some pseudoscientific statements - I admit pseudoscientific or even "Esoteric" bullshit can generally help some people with their "Faith" => "Actually following" (plus some placebo part) - I don't think this is something much applicable to more science-leaning people - (probably most people reading this)

@willssfx If your Business model is: - Let's play against Casino - think twice :) You can win of course. Unlikely to work long-term. Problem: - online "prop firm" has incentive for you to fail - because that's their business model - they own the rules - can change them, ban you..

@Alice_comfy Numerai has to have same bias problem. (but they are probably more sophisticated with selection) They are still around. But don't share public track record since this crash:

@Alice_comfy Yeah it's difficult to judge black box trading strategies purely from pnl curve. (Ok it's difficult even if you now logic, but it is at least something..)

@phoenixstealthy Exactly. There is often cheating involved in those cases (not only luck): - multiple accounts - reseting - negative skew strategies - not marking to market - martingale sizing, etc.

In practice. It's often much worse. Because things aren't that static like in this very basic "random normal" simulation. Often it's gameable: - you can get multiple accounts - bad results: Just reset it somehow - trade some negatively skewed strategy - (Volatility/Option selling.)