Quant trading economies of scale someone should give Doug his own blog

There are a few possible routes to establishing a monopoly in quant trading. Here’s one that seems to work really well. It’s kind of hard to explain, so bear with me. Many trading signals reliably predict prices, but not strongly enough to overcome transaction costs (i.e. exchange fees, clearing fees, liquidity costs, etc.). A stock can moves up or down $.01 in the next period, your trading signals predicts the right direction 60% of the time, and transaction costs average $.0025. Unfortunately after costs, you’ll end up losing $.0005 per trade, so it’s not a viable strategy.

But let’s say you’ve got three uncorrelated trading signals just like it. If you wait until all three point in the same direction, now there’s a 94% of the stock moving in your favor. You easily clear the transaction cost threshold and neatly make $.0063 per trade. When you gather together multiple uncorrelated signals, the whole becomes worth much more than the sum of the parts.

This is basically how a company like Renaissance Technologies operates. It has hundreds of people working in silo’d groups. Each group contributes to the overall fund’s strategy, but are largely unaware of what the others are doing. Alone any single group would probably not have a viable standalone strategy. That’s a big deterrent to people leaving, starting from scratch is really hard. And there’s only a few other major companies in Renaissance’s league, where they’re already strong enough to make money off of marginal signals. It’s a chicken-and-egg problem. To become viable in the space you need to accumulate a whole bunch of signals, but to attract a critical mass of talent with signals you need to already have a viable strategy running.

That one was “from the comments.


Why not let Doug do a few "guest posts" on MR?


"""Representing Aggregate Belief through the Competitive Equilibrium of a Securities Market """

Getting a blog is trivial; getting readers is the hard part..

I think you are onto something. Everyone I went to school with could come up with a large number of very profitable trades, whether at low, medium, or high frequency and code them in software. One challenge is making the ASICs and HW space for the HF HW and the bigger challenge is to convince someone to give you billions to trade with. If someone is stupid enough to do that you don't actually need to trade well to get rich. I don't now how I would go about convincing people to give me money, even if I can prove I can trade at least as well as any other firm.

I find quants as a mildly depressing case of market failure - all that talent, all those years of higher education in unrelated fields just to make miniscule corrections to market prices (under the charitable view, the uncharitable one is that they are glorified lottery players). Is it really worth it? Wouldn't it be better if they brought their formidable skills to bear on tough questions in physics or medical research or any other frontier of actually humanity-improving research?

Was it Glen Weyl that has a paper on taxation and talent allocation?

I would have loved to have been a physicist. But one look at the job market for academic physicists reveals that only a die-hard or masochist would seriously consider it. Good luck pursuing a career in physics and having any sort of modicum of stability or income before your late 40s. Nowadays getting a tenure track job in the field pretty much involves a massive stroke of good luck, or a lifestyle that totally precludes anything resembling a traditional middle-class American family.

Traditionally the Western university system offered a deal to the brightest young minds. Devote your life to scholarship, and we'll provide a comfortable, stable, minimal-work teaching position which will afford you with the breathing room and resources to think hard about the big questions. Once economic growth started slowing, this was one of the first priorities to get tossed off the sinking ship. There hasn't been any increase in tenured positions in decades, despite high population growth. Plus the rate of turnover has plummeted since professors are retiring later and later. Society has collectively decided that providing cushy administrative jobs, generous retirement packages and no-holds-barred healthcare for Baby Boomers trumps the importance of physics and other basic scientific research. So, why should I bend over backwards and live like a vagabond pauper to do "humanity-improving research"?

And medical research isn't much different. Nearly all the same critiques apply to academic research. And the pharmaceutical industry? Do you really think anything humanity-improving is being done there? Robin Hanson's quite instructive on this topic, medicine on the margin kills more lives than it saves. 90% of new drugs are no more effective than previous versions already on the market (see link below). Pharma research isn't about advancing medicine, it's about tweaking existing compounds to publish new patents and keep medical billing high. If medical research was actually trying to improve humanity, we wouldn't have had to wait 30 years since a single new antibiotic came out. What great societal returns, for "only" spending 15% of our GDP!

What do you do if you're clever and ambitious but born into The Great Stagnation? One good answer is that you play games. You take pride in honing your skills at your game of choice. This was a great point Peter Thiel made about why there were so many Soviet grandmasters. It wasn't that chess was any more compelling in the USSR, but there were so few opportunities to apply brainpower anywhere else. Well in 2016, one of the most interesting and lucrative games is called "global financial markets". Think about how many modern-day Federal regulations Tesla's workshop would be in violation of. I don't know if he would have been a quant trader. But I do know he would have found something else (adtech, bitcoin mining, competitive poker, frequent-flyer mile gaming, etc.) much easier to do in 2016 than actual engineering.


Thanks for your reply. Having left academia a few years after my Ph.D. after a hard look at my chances for financial stability, I agree with everything you say.

> Society has collectively decided that providing cushy administrative jobs, generous retirement packages and no-holds-barred healthcare for Baby Boomers trumps the importance of physics and other basic scientific research.

The fact that physicists who don't make it in academia often retire to finance, but not the other way around, suggests that academic physics isn't experience a brain-drain. This doesn't mean that the smartest finance guys aren't as smart as the smartest physicists; academia is selecting for masochism and luck as well as intelligence, so maybe the extremely smart but sane-and-not-crazy-lucky people end up in finance. But it does seem to suggest that academics gets to select from as high up along the IQ curve as we are able to measure.

So it doesn't seem to me that society is sacrificing physics research to save on physicist salaries. Rather, salaries have naturally been driven low by large supply, include important contributions from immigrants, while still putting upward pressure on skill. The fact that physics research has largely stagnated for decades is some combination of all the low-hanging fruit being plucked and (my opinion:) our inability to measure research output in a non-distortionary way. Under this second hypothesis, increased financial rewards to the physicists who survive the academic dog fight would *reduce* research progress (although this is my minority opinion).

Of the people I went to school with, one regarded as the smartest went into physics. It occurs to me that whatever their salary, it is low compared with the cost of a superconducting supercollider. So basically, they are paid in the form of massive equipment that nobody else gets to use. I suppose spending more would do better than spending on the welfare warfare state, but I agree salary is not the limiting factor.

Well said. It's really depressing, and certainly no way to run a railroad.

I find sanctimonious hectoring twits telling other free people how they should spend their careers mildly depressing.

You should be the social planner. Then all would be great.

And you should learn to read. I didn't say any of those things.

His 7.31 AM response was not to you.

Yes it was.

Yes, I had to slightly and logically extrapolate from what you said. And I was right. Wasn't hard. Bye.

It elicited a great response, interesting throughout.

Doug talks up the money you can make in quantitative finance, but Glassdoor says Renaissance pays in the same range as Google:



What gives? (You can click through the Google salaries to see bonus + equity, but it doesn't show bonuses for Renaissance.)

Also, my impression is that finance has contracted post-2008, so it's tougher to find a job as a new graduate, and now there's this new speed bump thing: http://www.economist.com/news/finance-and-economics/21701137-american-regulators-approve-controversial-new-stock-exchange-speed-bumps

SV seems like it has a more promising future.

>but it doesn’t show bonuses for Renaissance

Also keep in mind that working at rentec offers an enormous perk: you get to be part of the employee-only funds. How much is 35% per year on your networth worth?

I work at a competitor whose glassdoor-listed salaries are in the same range as rentech. From what I've seen, actual base salaries are a bit higher than listed, and bonus ranges from 100% to 2000%+ of base.

I've heard (unsubstantiated afaik) rumors that extremely high performers at google (besides founders, etc) also end up earning comp that can be one or even two orders of magnitude higher than typical starting salaries. My impression is that would be the 99.99th percentile at google though, whereas at a place like rentech it might be an 80th percentile outcome.

Didn't I read somewhere about a study showing the markets were efficient and that any signals you can pick up via machine learning (you, know, through quants) are fleeting? So this advantage lasts only until some other similarly massive player or players join together to duplicate it's success and the signals become too faint and/or too fleeting to catch with the existing technology?

Then on to quantum computing I guess -- if the algorithms realize enough benefit to justify the cost and the time it takes to process and load the data and...

Is the math in the example correct? I don't see where "let’s say you’ve got three uncorrelated trading signals just like it. If you wait until all three point in the same direction, now there’s a 94% of the stock moving in your favor" comes from.

Isn't Pr[ X = 1 | S1=1, S2=1, S3=1 ] = (0.5 * 0.6^3) / (0.5*0.6^3 + 0.5*0.4^3) ≈ 0.771, and not 0.94?

Let X = 1 if the stock moves up, -1 if it moves down, and assume these each occur with 0.50 probability. Conditional on X, each signal has 0.6 probability of agreeing with X, and 0.4 probability of disagreeing. We're assuming the three signals are conditionally independent given X.

Getting the probability up to 0.94 requires seven signals, not three.

I'm a total math amateur. Not even that. But you seem to be structuring this problem in a weird way. The 94% figure appears to come from the probability of all three signals being wrong equaling .064. Since they're uncorrelated, you can simply say .4 cubed is the probability that, taken together when they all agree, they will incorrectly predict what X actually does.

I posted a question about this at https://stats.stackexchange.com/questions/236718/trading-signals-example-from-marginal-revolution-blog -- I still think 1 - 0.4^3 is a mistake if you are trying to calculate Pr[stock goes up | all three binary signals say it will go up].

I think Adrian is correct. I'd explain it as follows: Let's say the stock is going to move up. The chance all three indicators will point up is .6^3 = 21.6%. The chance all three will point down is .4^3 = 6.4%. The chance all three will agree is the sum of these two cases, 21.6% + 6.4% = 28.0%. So, given that they agree, we can see that it's about 3 times more likely they all correctly point up. More precisely, they'll all correctly point up 21.6/28.0 = 77.1% of the time.
Adrian also made a good point at stackexchange that uncorrelated probably means uncorrelated after conditioning on the stock price movement.

Doug is FOS!

Competitive poker for Tesla?

It's no trick to make a lot of money if all you want is to make a lot of money.

He could short Tesla stock options in order to pay for his research. Or just convince SV investors that the grid was moving to DC, pay himself a million a year to do whatever he wanted, and let the investors sell the aeron chairs whenever the firm went insolvent.

Anyone got any papers on building ensemble models with such signals?

Maybe that is a strategy used by the organizers of https://www.quantopian.com ?

Or https://numer.ai/

You have to be really careful aggregating low sharpe ratio strategies like this. You ca data mine and find tons of them. You need to look at your meta strategy selection and evaluate with bootstrapping or other BS filters. Fischer Black had the idea outlined in the post at Goldman Sachs many years ago. The idea was that all the quants and trading desks would trade with a super trader on the top floor of the building avoiding transaction costs. However even in a sinlge firm it didn't work out that traders wanted to show their hands to other traders. This was way before the type of high speed trading that occurs now however. The stuff Renaiisance and others do can accurately be characterized as market making or liquidity provision I think.

I don't understand why its not more obvious to people that quants are doing market making. Spreads have fallen and quants dodge big moves from informed flow (which gets criticized as 'fake' liquidity, but really). They're making markets with more competitive spreads enabled by taking earnings from informed investors. It's interesting: informed investing is getting squeezed one side by savvier market makers and on the other end with fewer people putting their money in active management. What problems in public company performance might it lead to?

What's the economic benefit of such activity? Ray Dalio says it's the efficient allocation of a scarce resource, namely, capital. How does short term speculation on the movement of financial assets increase the efficient allocation of capital? Shouldn't capital be allocated to productive uses? Of course, I just don't understand how the economy works.

It's hilarious to hear these guys compare themselves to Tesla. Tesla died penniless. The Wright brothers ran a bicycle shop. Newton developed his work on the calculus, optics, and gravity at his family farmhouse because the university closed down due tot he plague. What the hell are these guys doing with their capital and time?

These guys are just clever rent-seekers, and they're going where the rents are. For a while, the rents were in the postwar Big Government Science racket. They're not would be Newtons and Wrights who'd be revolutionizing science and technology otherwise.

You forgot Einstein the patent clerk. This is a fair point, but just because they may not be all time or even generational talents doesn't mean they wouldn't move the ball forward. Plus today's progress tends to happen collectively rather than individually; we'll never know what impact taking XXX number out of the mix would have had.

"But let’s say you’ve got three uncorrelated trading signals just like it. If you wait until all three point in the same direction, now there’s a 94% of the stock moving in your favor. "

Well-known trade-off between selection bias (one signal) and over-fitting (many signals)

See: Novy-Marx, R. (2016). “Testing strategies based on multiple signals”, Working paper. Available at http://rnm.simon.rochester.edu/research/MSES.pdf

The problem with trading strategies is not the signals because they are trivial but changing market conditions. If you know when market conditions change you can switch between trend-following and mean-reversion with simple systems. See my paper for specific examples (plug here) Limitations of Quantitative Claims About Trading Strategy Evaluation


Honestly speaking, I really love this blog with so many interesting stuff been put up regularly, I love reading all that and been a Forex trader, I get plenty of help and ideas. I am lucky enough to be working with a broker like OctaFX which too helps me greatly to do with their rocking features with 50% bonus on deposit, rebates, instant payments and many such things, it’s all absolutely picture perfect and ideal way for me to work with complete comfort.

Alpha Combination

It today’s markets, rarely is any single alpha significant enough to be the sole basis of an investment strategy. A successful strategy usually includes many individual alphas; if they are strong enough, a few can suffice. The goal at this stage is to implement a weighting scheme which takes as input many normalized alphas and produces a single alpha which is more predictive than the best individual alpha. The weighting scheme can be quite simple: sometimes just adding ranks or averaging your alphas can be an effective solution. In fact, one popular model does just that to combine two alphas. For increased complexity, classic portfolio theory can help you; for example, try solving for the weights such that the final combined alpha has the lowest possible variance. Lastly, modern machine learning techniques can capture complex relationships between alphas. Translating your alphas into features and feeding these into a machine learning classifier is a popular vein of research 4,5.

Source: https://blog.quantopian.com/a-professional-quant-equity-workflow/

Comments for this post are closed