Point72 seeds Quantopian-backed fund with up to $250mm

[deeds]

thanks, sigma



second your point...would go further, in that, given march of time and reflexivity of markets relying simply on backtest information will not be adequate...further...could it be that fully "optimizing" with respect to back test is...sub-optimal...without taking into account online nature of real task at hand, or estimation parameters...seems shoe-horned into typical inferential statistics, made for bayes

[sigma]

Thanks, deeds



That is my concern for seeding these types of strategies with outside money: the imposed risk limits can be too tight for a strategy to survive in the long run and produce any benefits for its contributor. Also, given the fact that most likely you cannot adjust it fast without an external approval when it is live.



Does anyone have such an experience / comments?

[rftx713]

>As a result, even if you want to target a specified level of the volatility, you need to have your algo running for some time to estimate its vol. What do you think?



I think that makes a lot of sense (as do the additional comments)... add it to the list of why I'm so confused about how they actually make money.



From reading this thread it's clear I'm the amateur in the conversation, but I would still want to know about the size and structure of the fund overall more than anything. My logic being: who cares if it's poorly run if its existence is to serve a marketing function more than a PNL one? Or maybe, what is the fee structure on this outside capital? Are they just trying to hang on and collect fees and not blow up, while the fund also serves a marketing function for it's sales platform? I'm lost.

[sigma]

I think that the quantopian aim ultimately to become a fund-of-fund type of allocator to quant strategies. For sure they will reap benefits if they get a share of a typical 5% allocation to alternatives by large institutionals.



However, if there is a definition of a black box, this is the top of it.

As an investor in such funds, I think there are three primary type of risks:



1) Performance of individual quant strategies. Crucial is independent back-testing before putting them live.



2) Allocation procedure to best performing (in- and out-of-sample) strategies. Of course, they will aim to select 99.99% of all contributed strategies, but from the risk point of view, each strategy should not get no more than 5% of total funds. Then comes diversification among strategies.

First, they don't want to be concentrated to any particular type of strategy, say momentum.

Second, they need to stay market -neutral across aggregated positions from all strategies.

Obviously, their task is simplified if they have N uncorrelated strategies with relatively constant volatility, but I suspect it won't be the case.

Then of course it is important to account for the dynamic nature of exposures and volatilities of selected strategies



3) Operational risk. Important is:

a) How is a strategy selected to the live basket. Does it get a diversification benefit?

b) How stop-losses are triggered? Is there diversification between drawdown times?

c) How a strategy is removed once its loses its edge and how new strategy is introduced?



My biggest concerns is that these points (especially 2 and 3) cannot be really back-tested given the discretionary nature of decision making process. As a result, only after the "fund-of-strategies" is up and running live for some time we can get some understanding of its risk-reward potential.



When you ran a basket of strategies in your fund or account, you should think of them in terms of the total performance of your fund. Here, I may see the separation of interests between a strategy developer and "fund-of-strategies" manager.



Finally, there is the diminishing alpha effect. If this "fund-of-strategies" is indeed able to generate alpha, you want to be early in the game before the crowding will reduce the alpha



From Point72's standpoint it must be that they want be the first in the game in case it is indeed worthwhile. In the end, 250mm is not that big money for them to pay for staying in the first place if there will be indeed any potential allocation to these type of "fund-of-strategies" by large institutionals. If this "fund-of-strategies" will indeed prove to be a success at the initial stage, the subsequent management fees will pay off handsomely.

[goldorak]

I think there is a word to summarize all this quantopian stuff: "hype".



Strategies are nothing else than filters applied to time series. Developing filters on time series is already difficult as hell, but being able to select the right strategies (which is nothing but filtering out the filters) makes you evolve in a totally different league. I would even claim this is not the same sport anymore. You could be the best 100m runner, but still that would not make you the best 100m swimmer if you trying applying the same exact technique. Hey dude, you may think of staying above the water rather than trying running on the swimming pool's ground as fast as you can? If you want to get a chance in that different sport, you better stop thinking like institutions are thinking now. sigma is not wrong in his detailed comment, but it is just not enough. I am not going to extend too much on this topic as I consider further knowledge competitive advantage.



However, when it comes to quantopian, although they are not the worst you could imagine, they are still a number of wars behind. Their discourse sounds like the conquistadors coming to America and facing a modern end of 20th century warfare. For example, their recent paper [gold], although being a lot more advanced than whatever FoHF bs you can find out there, is just laughable at best. To add insult to injury, they actually MAKE PUBLIC criteria of interest, and of course encourage developers to concentrate in overfitting their criteria.



No, definitely, they are not up to the level needed to do what they intend doing. Hype.

[Nonius]

Tough year Goldorak?

[goldorak]

Terrible year, but nothing uncommon. At least we don't need to establish an investment committee who will name a sub-advisory committee who will issue a recommendation report who will be sent to a Board whose decision will anyway be in his own interest. The report would of course cite a number of high level analysis that show that had we made things differently over the last 6 months we would have made money and we should now apply this methodology rather than the one we chose a long time ago based on the same criteria (I mean mistakes).



The algos will discard, give chance to diversity, stay open to new opportunities on a day to day basis. The binary aspect of ex-post unplanned decision is over. How many HF out there did you see just hanging to their superb strategy for years while slowly bleeding? Too many I guess. You should note this is not all about the decision process. If none of the filters you provide is unable to extract any temporary piece of information out of the time series in the analysis, the decision process is not going to invent performance, excuse me, ALPHA as the quant hype loves to name it.



Quantopian is just doomed to fail. Economically the hype may prove to be temporarily rewarding though.

[cf_mstr]

They claim your algos are kept secret, I claim they save all algos information and sell them/do whatever they want with it. A profitable algo is worth a large chunk of cash I guess... And now they get thousands of members developing/testing them, at discount.

[rftx713]

In response to the above, I suppose goldorak's point would be: what even is a "profitable" algo, given there are various implementations, effects on the overall portfolio, etc.? Further, why would they be better at taking algos and implementing them? I can have a million golden retrievers building my house; it still won't be as high quality as 10 professional builders. Further, choosing the best 100 golden retrievers still won't produce equal results.



Or did I totally miss everyone's point? (Sorry for the weird analogy.)

[goldorak]

Partly true. In a bayesian approach to the problem, you start with a complete agnostic view on who is a golden retriever and who is a professional builder. It is with your experience on their track record that you will adapt and decide by how much you are going to trust them or not. You do not care entrusting 1 or 2 golden retrievers by mistake. What you want is to find the quality builders with a limited number of tries. Once identified, you need to quickly identify the quality builders turned golden retrievers.