Back to Community
Point72

Been reading about Stevie Wonder and his new CEO. Some very good articles out there which fit in very well with my views on basing ML and Quant stuff on fundamental research, or at least theories about what makes a good stock. About combining the two.

Yes Steve, all very plausible, frankly.

My scepticism remains (think Naseem Taleb and randomness) and I believe more firmly than ever that the only "safe" way to make above market returns is to play to some advantage rather than using a probabilistic approach. As with HFT (b/o spread, front running, spoofing) or in my own case with hot IPOs in the good old tech driven 90s.

Long / short ungeared maybe. 6 times leverage. Ouch...eventually. But what do I know....

In any event, a valiant effort.

14 responses

in my own case with hot IPOs in the good old tech driven 90s

What was your advantage? Or did you just get lucky?

The simple advantages were as follows: I knew, roughly, whether the issues were hot. I only went for us issues with bulge bracket underwriters who usually held the bid at or above the issue price at the open. I sold at the open on the day the stock started trading.

Oh - and of course I was subscribing direct with the underwriters and got the stock at the issur price.

Haven't played the game in a while but it became a lot more difficult to get stock outside the US.

Long / short ungeared maybe. 6 times leverage. Ouch...eventually. But what do I know....
In any event, a valiant effort.

The argument is that somehow it works, that there is no survivorship bias, that whatever Q has in mind will work, based on the fact that it is established within the industry that it is a path that can be made to work. But then have a look at:

http://news.morningstar.com/fund-category-returns/long-short-equity/$FOCA$LO.aspx

At a gross level, the case doesn't seem very strong. Also, if one sorts on the '5 Year(%)` column, there are a lot of newcomers. Either it is a growth industry, or there is a lot of turnover, due to failures.

Regarding leverage, it is a total mystery to me. I guess the basic formula is that Q would demonstrate (1+r)^n consistent returns, suggestive of a bank CD, and then they'd get some entity to lend them money, to increase the capital level (presumably, the broker makes the loan, but where does the broker get the money?). If things start to deviate too much from the (1+r)^n return, then the lender pulls back his capital. I don't understand this whole thing. If I "lend" $100 to Q for their fund how would this be different from my "investing" $100? I guess if I'm an investor, there is friction preventing me from pulling my money on a dynamic basis, whereas if I'm a lender, as part of the loan agreement, I'd have my hooks into the minute-by-minute status of my portion of the investment, and could add/drop capital per some agreed-upon rules, but since I get to do this, I don't get any of the return stream, I just get to charge fees/interest. Correct?

So, why would 6X leverage be a problem for Q? Would it tend to endanger their 1X capital outlay? I guess the idea is that a "margin call" would be made at an inopportune time? Wouldn't it be the lender who is at risk? Or is the idea that they'd grow the business based on returns from 6X leverage (e.g. headcount, leases, pc's, desks, cubicles, etc.), but if the leverage gets reduced, then they won't be able to pay their bills?

Ha ha yes I also looked at long short equity funds. Re disaster, despite what the talented Mr Fleury might think, black swans usually eventually f**** those who over use it.

Read Nasim Taken....very amusing stuff and his views seem to be born out over time. Over an over again throughout history leverage has usually put a spoke in the wheels.

Survivorship bias probably accounts for those who have not yet blown up.

The usual HF model is to make a packet out of fees and close the fund when it craps out. Then start another one.

There was a lot of fanfare around the announcement of the Quantopian/Point72 relationship formation. They invested $2 M through Point72 Ventures and subject to undisclosed milestones, will start to invest up to $250 M in users' algos. My understanding is that at this point, Quantopian has funded 15 algos, at an average of $100 K each (~ $1.5 M total). Presumably, aside from the accounting, and legal mechanics, there are milestones tied to the consistent real-money performance of said algos, relative to their backtest and simulated live trading periods. I guess it all depends on how much stake one can put into the backtest and simulated live trading data. It seems ideally one would want 2-3 years of real-money data at a relatively low level of capital, before ramping up.

I've tried various angles to see if anyone at Quantopian understands why the multi-factor, long-short approach outlined in the workflow might work, but the argument tends to be "Grant, you are an idiot. This is what everyone does. There are billions invested in this fashion. It'll work, just like it has worked for everyone else, except better." It is called confirmation bias, I think. I'm still totally mystified by the mechanics of a large number of factors combined via computational magic to produce (1+r)^n type returns over decades. It is hard to grasp, intuitively. So, I ask, "Hey, are there any examples out there? Could you show me an existence proof, by running a backtest on Quantopian?" Etc. Reasonable questions, no? Presumably, evidence exists. I've even asked if anyone at Quantopian has seen one of these things first-hand, in action--no response. Very mysterious.

I know it is dull but you might find it helpful to look at hedge funds in general and what they have / have not achieved over the years. I am not knocking Q (far from it) but I do think Nasim Taleb and others have a good point on randomness and survivorship bias.

To quote but a single example - Atradis, once Singapore's biggest hedge fund. The founders walked off with over $100m in fees apparently before they closed the fund down. Apparently they deemed their model no longer profitable.

I can't help wondering whether highly geared long short is picking dimes in front of a steam roller.

Each year some new operation discovers the holy grail...and then investors discover it isn't.

It isn't the long short I quibble over, its the 6x gearing. I would need to study the history of L/S funds to get a really good grip but of course hedge funds are reticent to provide much if any real information.

I think Q are doing a great job on many fronts and I take my hat off to them. I am myself working hard on ML and will post some algos when I am further along the line. I am always prepared to keep an open mind. My cynicism may yet prove unjustified.

Simple quant strategies used to work when there were enough anxious irrational discretionary players in the market. Lately there's a bubble in rationality. Investors have transitioned from full-service brokers to RIAs or robo-advisors, stock picking to indexing, mutual funds to ETFs, etc. If you look at any properly constructed backtest, you see diminishing returns in the past few years. Trying to solve that problem with leverage justified by illusory diversification seems indeed to be a risky path. Once everyone is a quant, there will be no market.

Very good point. I used Kaufman's Efficiency Ratio to look at the alarming decline in trend efficiency in the futures markets over the last 40 years. My findings are given expression by the collapse of JWH and Bill Dunn after decades of successful performance.

I gather that the recipe for a "geared" long-short multi-factor is to mix together a bunch of factors to get a relatively small but consistent return (say 2%-3% per year above inflation, SR ~ 2) and then to borrow money (i.e. "leverage" which sounds like a pry bar, that works every time). With 6X, the effective return becomes more like 10%-15%. Then, Point72 gets X% and Quantopian gets the rest as incentive, and to keep their business going. And I suppose that a family office like Point72 might get favorable borrowing terms? Perhaps the broker pools the risk?

@ mhp -

If you look at any properly constructed backtest, you see diminishing returns in the past few years.

Do you have a particular backtest in mind? This seems like a hard way to prove in a convincing way that there is a secular trend, given how easy it is to bias/overfit backtests.

@ Anthony -

I think Q are doing a great job on many fronts and I take my hat off to them.

I agree at some level. They have attracted some really bright employees. And at some point, I was not able to keep up with the increasing sophistication of their platform. The flip side is that the main players have all but stopped interacting with users on substantive issues, let alone encouraging other stakeholders (e.g. investors & Point72 folks) to chime in (although perhaps they do in ways other than this forum). The insularity is bad. And I don't think the principals ever thought hard about what a radical, egalitarian, crowd-sourced effort would look like (or maybe they did, and just couldn't see how to pull it off and sell it to investors). It has to be structural, I think. Users have to have a path to owning a stake and having a say. At a high level, it seems to be business as usual, with the ultra-wealthy exerting power over the masses. Quantopian is missing an opportunity, in my opinion, but what do I know.

It would be interesting to see some high-level metrics of the amount of "slack" or inefficiency in the equity market. Here's one example:

http://homepage.sns.it/marmi/lezioni/talk_risso.pdf

Interesting comment in there: "Once the DotCom bubble burst, investors purchased real estate which many believed to be more reliable investment." We know how that turned out.

Maybe there is some direct evidence that at a high level, equity long-short strategies have been getting harder to implement versus time? The workflow concept is wonderful, but if there is a strong head wind, then it may not be the best strategy. Or perhaps the gross effect is counter-balanced by technological advances in data sources and analysis tools (combined with global crowd-sourcing/talent recruiting, which would seem to be the Quantopian play)?

@Anthony: Dunn's programs made new all time highs in 2016. Not sure what you mean by collapse?

70% dd at one stage and effectively he and his son ceded management...from what I gather having met them in London a few years back