What happened to acrary?

T

test_user

I recall reading acrary's posts around mid 2000s. He made a great impression on me back then.

Lately I was looking back at it and I can't help but think that not everything is as perfect as it seemed.

For example the edge test which is supposed to prevent overfitting. It's a p-value against random results. It's very easy to overfit - by definition a certain fraction of tested systems will pass the randomness test simply by random chance. It's a basic multiple comparisons problem.

Next there was the famous thread "System Development with acrary". It could be summarised as optimisation of portfolio weights for maximum Sharpe. His method was an exhaustive iteration through all weight combinations. However completely unnecessary as there are analytical solutions that don't require exhaustive search, namely Markovitz optimisation - stuff that's been known since the 1950s. It's yet another exercise of naive overfitting, since Sharpe ratio estimations typically have high uncertainty. In many cases an equal weights portfolio outperforms naive optimisations.

Lastly, the alleged performance. I checked the thread, there's an example of 5 trading systems portfolio (4 for ES and 1 for NQ) that shows Sharpe ratio of 1.61... monthly! That's approximately Sharpe ratio of over 5 annualised. Are we supposed to believe that it is achievable with a few day trading systems on a couple of index futures markets? It is simply unrealistic. There are real world cases having annual Sharpe ratios of 2-3 (excluding HFTs and option selling blow up artists). I can think of a couple, one is the famous Renaissance Technologies certainly trading more than a couple of index futures markets and the other Ed Thorp's Princeton Newport Partners running statistical arbitrage on thousands of stocks.

Let's assume the performance was true after all. With this kind of ability one would be at least a billionaire in short order. We would very likely know about this, yet there's nothing out there.

A secret billionaire or a fake guru?
 
Wow, this is going back a long time. I'll have to go back and re-read the posts. I'll see your acary and raise you an abogden / MAESTRO.
 
I think acrary was legit.

supposedly he worked at bear stearns in their stat arb desk. I knew people on that stat arb team from the late 2000s and it looked like a legit operation.

I have seen his posts on other forums and they always seemed sincere and he never sold anything.

I can’t speak to your math but one thing he said that stuck with me was that it was easier to find simple strategies that earn in specific environments and then focus your efforts in determining which environment you are in. That could be the source of his high sharp.
 
newwurldmn said:
I can’t speak to your math but one thing he said that stuck with me was that it was easier to find simple strategies that earn in specific environments and then focus your efforts in determining which environment you are in. That could be the source of his high sharp.
More...

That sounds easy on paper but I have yet to see any evidence that it is possible. For example, in real life by the time you determine that the market is trending it is too late to hop on that trend. If you don't try to predict what environment you are in and simply combine multiple systems, it's still nowhere near Sharpe ratio >5. A realistic Sharpe ratio for such a set up is ~1. Of course in a backtest you can get anything you like.
 
test_user said:
That sounds easy on paper but I have yet to see any evidence that it is possible. For example, in real life by the time you determine that the market is trending it is too late to hop on that trend. If you don't try to predict what environment you are in and simply combine multiple systems, it's still nowhere near Sharpe ratio >5. A realistic Sharpe ratio for such a set up is ~1. Of course in a backtest you can get anything you like.
More...

I can’t speak to any of that.
 
test_user said:
That sounds easy on paper but I have yet to see any evidence that it is possible. For example, in real life by the time you determine that the market is trending it is too late to hop on that trend. If you don't try to predict what environment you are in and simply combine multiple systems, it's still nowhere near Sharpe ratio >5. A realistic Sharpe ratio for such a set up is ~1. Of course in a backtest you can get anything you like.
More...

IMHO, I wouldn't accept those results on their face but I also wouldn't be as skeptical.
 
I 'member MAESTRO too. I didn't follow his posts closely but I recall something about his secret formula where he takes the price, divides it by ATR then makes dollar inflation adjustment (or something like that). This supposedly transforms the price data into normal distribution and for some reason it becomes "predictable", whatever that means.
 
You must log in or register to reply here.
Back
Top