* 2013 - 5% over
* 2014 - 15% under (and negative overall)
* 2015 - 30% over
* 2016 - 10% over
* 2017 - About even
The other problem is scale - this is pretty evidently an advertising attempt in order to raise more cash. It's a lot easier to return 20% on, say, $100M AUM than it is on $1B AUM.
To be fair -- these caveats are true for almost every fund you hear about. I do think that's kind of my big issue with the article though, that this fund is just like every other fund; it's intrinsically an actively managed portfolio with the same shortcomings as any other. It's just got a good publicist!
Unless there is a way to break this data down much more finely (by 6-day window, by individual asset) in which case it's possible there's a high p-value but low-power effect. Even so, if there are more than 100 experimental firms, a single one with p-value .01 isn't evidence enough to jumping to big conclusions.
Maybe you'd weight it based on class size? 4 left isn't bad in a class of 100, but it's a bit high in a class of 10.
"Fuzzing" for alpha that way would be really inefficient.
If that's all you're doing, you might as well go all-out on claiming to have secret sauce that produces your alpha. I'm not expecting you'll get any, just that you might as well have a media-friendly way of picking if you don't have an actually useful one.
How do you define an "attempt", and how many "attempts" do you need to beat the market (consistently) by chance?
If you're going to quantify survivorship bias, you can't use entire years as data points, because that doesn't properly represent the amount of activity that occurs. We should reason about each event, because if consistency emerges on an event basis we might not even need more than one year for our sample. The decision-making that is being empirically examined here (i.e. acumen capable of beating the market beyond chance) ostensibly functions on trading events, which means years are not the correct data point to use (and will provide an incorrectly pessimistic sample).
If the data is reported on a yearly basis then it's pretty much the same thing.
I can quibble about the odds of each individual trade resulting in profit or less being binary, but for the sake of argument it'll do. But a 50% chance of beating the market each year isn't supported by anything.
The grouping of data reporting doesn't suggest anything about the underlying data if it doesn't also share the same probability distribution. The trades are the events which determine if a fund will outperform on an annual basis, and we can group those trades by day, week, month, year, etc.
Strategy B: buy two random stocks.
If the expected payoffs were mathematically different then you'd have an arbitrage opportunity. Then apply induction.
Wow, so they are ~27% off the mean!
The part that isn't clear is how this compared to typical performance variation - then we can run our p-tests. But I suspect its pretty good, given the news worthiness of the article.
Also for the actual return:
2017 +10% (til June)
Or for the 4 whole years:
(1.22 * 0.98 * 1.33 * 1.22) ^ 0.25 = 1.18
I'm holding out for a 4-way contest of quant, horse, monkey, and algae. I won't be betting on the winner, either.
That's perhaps the most annoying thing about "machine learning" in general and heuristic algorithms in particular.
These researchers hack together some rules to put together an algorithm, and only after that does the real work start: to come up with a metaphor to explain their model based on some clever story on why they decided to generate random sample points or filter training points.
In the end, the whole field starts to look like a bullshitter's ball, where everyone tries to one-up each other with the biggest bullshit metaphor to sell their an algorithm which is actually only a very minor tweak on an established age-old concept.
"We use particle swarms to generate new solutions, which are then genetically modified and subjected to a darwinian-inspired differential-evolution filter, who are then analised based on the behavior manifested by wolfpacks to search and hunt for their prey, and whose sub-optimal solutions are eliminated by following nature's resource-exhaustion megakill phenomena."
Bullshit all around, but it sells.
There are skeptics, too. Emanuel Derman, who was among the first physicists to work on Wall Street, doubts that biologists possess secret sauce for investing. Derman rose to lead the quant risk strategies group in his 17 years at Goldman Sachs Group Inc. He found that as physicists applied their expertise of the laws of motion, atoms and mathematics to investing, their models didn’t work nearly as well as they did in a lab.
Newton’s law of gravity hasn’t changed for eons, Derman said, but human behavior in markets changes all the time, wreaking havoc on even the best models made by scientists.
“I’ve developed a lot of skepticism about anyone bringing their expertise from one field to another,” said Derman, author of the book “Models.Behaving.Badly” and a Columbia University professor of financial engineering. “They say stocks are like atoms, or like genes. But stocks are not atoms or genes. There is a resemblance, but ultimately they are very different.”
Also 'the map is not the territory', all models will be unable to deal with all possible behaviours of the reality they are dealing with in a correct way.
Your first example seems like it maps well to that cognitive bias - someone assumes that a chess grandmaster has catch-all capability because they demonstrated expertise in one area, and then they flare out in an orthogonal area.
Mathematics and engineering (at least, computer science) overlap in many areas. If someone excels at mathematics it doesn't prove that they'd be good at programming, but I'd bet a significant amount of money that the mean mathematics major is more capable of programming than the mean population in general. If that holds, there's no bias in trying to cross-pollinate expertise, even if it ultimately doesn't work out.
A lot of the work that occurs in finance is legitimate mathematics and has very close ties to physics. The heat equation is directly used in Black-Scholes; securities can be modeled as stochastic processes, which means that much of the models that apply to Brownian motion also apply to them. Outside of quantitative derivatives pricing (and more recently), hedge funds can apply the same scientific computing techniques used by physicists and computational biologists to analyze vast amounts of data (more than they know what to do with).
2. That bet is paraded more than it should be. Buffett bet against a "fund of funds", which is the aggregate performance of the industry. We didn't need a decade long bet to tell us most hedge funds are a poor return of capital, just as we don't need a bet to realize that the greatest n participants in many fields are mediocre.
I would have gladly taken that bet with Buffett and won if I could have chosen a single firm. But Buffett wouldn't have taken that bet, because (to his credit) he understands this point already as a savvy investor. The bet proves that the industry overall is mediocre, but it says nothing about the top firms that mostly don't even take outside capital anymore because they're so successful.
But I think you are downplaying the significance of the bet. If it was an obvious wash like you are making it seem then why was the bet even placed? What top firms are you talking about?
2. Firms like Baupost or RenTec would have handily won that bet against Buffett. But like I said, Buffett wouldn't have made that bet, because Buffett is a smart better and already knows all of this. Buffett has never argued that the market is perfectly efficient and resistant to alpha harvesting; in fact, he has publicly taken the opposite position in letters to shareholders.
This is what 99.99% of consistently profitable quants do. It's boring, unexciting, and VERY profitable. The crux is...you must have that low latency connection to the exchange, and you must have priority routing for your orders. Bonus point if you have market maker status(but if you have that, why are you doing this in the first place?).
The deep learning craze seems to be infiltrating the minds of a few algorithmic funds, but it doesn't stay long(either they stop trying it, or they blow up their fund). Positions of any reasonable size(such as that required to move the market) are opened by human beings. Human beings operate on emotion and mob mentality in the market, so that is what you capitalize on when designing your algorithm.
Disclaimer: I'm looking at this with an ultra short term timeframe, such as 5-15 seconds being the maximum time in market per position. If these guys are targeting longer term positions, then all bets are off. The algorithms I create and maintain work in this timeframe, and the majority of my competitors algorithms are in the same timeframe.
This is entirely incorrect. Quants do not, as a rule, run liquidity providing strategies with market orders. Market orders pay for liquidity; they are the buyers for what HFT is selling.
> said Lun, whose computer holds trades for an average of six days
Does this exclude outliers? What's the median duration?
Another disclaimer, the markets I'm quoting these averages from are the ES and ZB.
In general, people have a poor understanding of how to evaluate an investment manager. It's not enough to just look at absolute returns and compare them to the S&P, you need to correct for market exposure (the beta). Even then, it is not that straightforward: this is one of the best overviews I've seen (the author of the blog, Robert Frey, was a former managing director at Renaissance Technologies, the most successful hedge fund of all time)
To make the "correcting for exposure" aspect concrete, suppose you have the opportunity to invest in a poker player that generates a 10% return on capital per year. It wouldn't really make sense to compare this return to the S&P 500 returns, because the beta is very close to 0.
Well he is certainly surrounded by some impressive people.
I think at this point in the search for alpha, wall street has employed applied mathematicians, physicists, code breakers, engineers, economists, chemists, computer scientists, sociologists, and now computational biologists.
Each one of them brought some new/novel mathematical techniques to the field. Do Medieval Historians learn any specialized math?
Maybe he can beat the market reliably but he's only managing $20 million. The big question every quant fund asks is can this strategy provide alpha at a salable level of investment.
A 3 year track record is plenty long enough to prove out a system and provide a track record. It's a troubling sign that there is only $20 million in his fund if.
I'm curious as to why you say a 3 year track record is long enough to prove a system. I don't necessarily disagree (though I think number of trades executed in that timespan and the type of trading strategy might be as important as the timespan itself), but I'm interested in your reasoning.
It's important to note that 3 years doesn't mean 3 data points. It really depends on the funds average trade horizon. Which is, I think, exactly what you were referring to.
An HFT firm trades at such small scales that it can use its daily returns such that each year actually provides 252 data points.
On the other side of the coin, Berkshire Hathaway would need benchmark times longer than a single year.
I'm assuming the fund has holding times of around a week based on intuition and prior knowledge of alto of different fund investment structures.
The thing to understand about hedge funds is that most of them change investment strategies at some point in their lifetime such that historical records no longer really apply. This can happen for a number of reasons:
1) markets get crowded and force people to search for alpha somewhere else
2) funds get larger and existing strategies don't have the capacity to manage the new money.
3) traders leave and new traders have new ideas.
3 year is an industry goldilocks mark for comparing hedge funds. Not too long to take into account old strategies that are no longer employed and not too short that it doesn't allow the strategies to play out.
I cant' remember the exact number but Victor Haghani of LTCM fame talked about this and said it would be something like 143 years of data to know if a biased coin that comes up heads 60% of the time is biased to a 95% confidence level.
Obviously this isn't workable and as such we have to use smaller time frames.
No information about what these models are is given. But it seems more "I created a system which can predict cell changes and stock market movements" than "I used Algae to predict the stock market".
> Lun, who was born in Hong Kong, splits his time between his firm in Pennsylvania and lab at Rutgers, where he’s undertaken an ambitious long-term project: creating computer models that predict how cells behave, using data from blue-green algae and other sources. The models allow Lun to re-engineer genes for useful purposes: he has modified E. coli for production of bio-fuel for transportation.
If the bet turns sour, you close the fund. The idea is that you grow the fund and take your two and twenty while your bets are winning, and you keep your prior years' two and twenty after your fund collapses.
Lets see how it performs longer term (10 year period).
* Not everything financial is zero-sum, but this sounds like it is.
Year Market Algae
2013 +16% +22%
2014 +14% -2%
2015 +2% +33%
2016 +12% +22%
2017* +9% +10%
I'm reminded of Linus Pauling: He made amazing, fundamental breakthroughs in chemistry and quantum physics, but when he applied his genius to medicine, we got orthomolecular medicine and mega-dose vitamin C as a cure-all, something which has been roundly disproven by actual evidence.
That said, investing in algae could be a good idea. It has potential as a cheap, high-volume input to synthetic food production.
For the last year I've only been a glorified webdeveloper working for molecular neurobiologists at the Karolinska Institute, but from what I understand it is all about untangling vast quantities of high-dimensional data: data sets of tens to hundreds of thousands of individuals cells, where for each cell the expression levels of tens of thousands of genes are being measured (in what stage of development in which tissue was the cell harvested).
If you can find algorithms that somehow make sense of how these cell populations and genes interact and develop over time, I think it is not out of the question that the same algorithms could make some sense of the aggregate behaviour of the stock market, give a decent data set as input of course. Especially given that most of these algorithms are forms of machine learning, so don't necessarily require an a priori model of what is happening (I mean, if I understand correctly, uncovering that model is precisely what the biologists are after).
Where can I sign up to gamble with other peoples money with my black-box AI algorithms based on fungal evolutionary genomics?
“It’s hard to explain simply why and how it works"
I also did not understand why cells specifically provide such good insights for markets, as compared to any other complex natural system (like weather or ecological systems).
Find non-diversified asset pools based on mutual information of prices. Avoid too much exposure to any pocket, but do arbitrage within a pocket.
If we all planted trees and algae then we'd increase carbon sinks and solve the greenhouse gas problem in a sustainable way.
I am surprised more money - or at least ink - isn't funneled into this!
 - https://www.google.com/finance?chdnp=0&chdd=0&chds=0&chdv=1&...
 - https://www.google.com/finance?chdnp=0&chdd=0&chds=0&chdv=1&...
In the graph hedge funds have performed very badly against the S&P. Given the massive fees a lot of hedge funds charge: what is the incentive?
Also, not sure about hedge funds, but alot of mutual funds exist with a specific focus - one country, sector, asset class, etc. They try to do well within that category...
Just because the 'average' does worse, doesn't mean there aren't a few that do far better if you pick the right one..
Whether you think it is good to invest in that sector over the short or long run and why is up to you based on your portfolio and expectations for the market over the holding period..
As others have said - diversification.
The number of data points is important when looking for trends, or for cleanness of data. However, they're not showing us the data. It could be hugely volatile, or fairly linear.
1. Options/Futures etc by Hull