@Grant,

This is a great question and perhaps Q will not disclose as this may be in the realm of their proprietary rights. Let me try to attempt to answer in a perhaps a very simplistic way. Given say 20 individual raw signals, one can start with an equally weighted baseline ran through an optimization with some constraints to establish a baseline. Next run the signal combination through the optimization that adjusts individual weights and see if it beats the equally weighted baseline.

@ James -

Something like that maybe. The thing is, only a simple linear combination of factors has relative weights (e.g. sum of z-scores) and the optimizer can jumble things. The formula of (signal weight in combination) x (total fund return) doesn't mean anything unless one knows how the weight is computed and how frequently. It also doesn’t allow the author to have any idea what his signal is worth. Under the old system presumably authors could estimate their individual returns and make sure they are paid the agreed-upon percentage.

The formula of (signal weight in combination) x (total fund return) does(sic) mean anything unless one knows how the weight is computed and how frequently

Absolutely, it is somewhat opaque but presumbably it is why fund authors get 10% and Q gets the other 10% of fund profits. Let's call it division of labor or expertise. Fund authors' tasks ends at providing good raw signals, then Q investment team takes over signal combination, portfolio construction and execution. Mutual trust, I guess.

It also doesn’t allow the author to have any idea what his signal is worth.

Say Q re optimizes signal weights every end of month and executes these weights for the next month, one can track the weight given to his/her signal (which should be equivalent to "signal worth" for that given month) vis a vis of total fund performance. The process of determining the weights is what is somewhat opaque.

No big deal here. I just think that presenting a formula that can't be interpreted without more information is bad practice. In the end, the more I try to think about it, the more I realize it says nothing, other than there is a linear relationship of payouts to the total fund return--well, there ought to be! I guess under the constraint that the weights sum to 10% of the fund profits, then the weights are just a way of sharing the love across the authors, and perhaps are just negotiated individually, and aren't a kind of real-time output of any algorithm, which would be fine, too. But why not provide a few more details?

Another factor would be the confidence level in the signal. For example, a signal that has had only 6 months of out-of-sample data, and no real money trading, versus one that has been running with real money for several years are two very different beasts. Nobody trusts a backtest, and especially when the details of the strategy are opaque (in the Q case, the code is not available for review and if used, the outside data being fed in are of unknown origin). It's kinda blind faith (unless the author shares his code, which should have the effect of bumping up his weight, since the risk should go down for the investor).

I certainly don't envy the Q team in this regard, trying to set up a totally black-box hedge fund (has anyone ever done this?). Even with full alpha factor/signal transparency, it probably takes at least a couple years of real-money trading to have any sense if things are real (really, probably umpteen market cycles...but who has the patience for that). The Q fund is mighty audacious...wonder if the risk is worth the potential reward?