Interview with a Hedge Fund Manager: "I'm sure today I would never get hired." | Hacker News Reader

Interview with a Hedge Fund Manager: "I'm sure today I would never get hired."

nplusonemag.com

55 pointsgarret21 comments

The "Wouldn't get hired" comment is pretty common -- in fact, an interview with Daniel Loeb about his multi-billion dollar fund ended with him saying something like "I couldn't get hired here. Bad grades in college, no MBA. If you want to get a job here, you have to start preparing when you're thirteen." (So once again, these guys are all using the same model).

Some funds have a program explicitly designed to hire people who 'wouldn't get hired' under normal circumstances: http://deshaw.com/articles/Alpha_2.pdf .

"I think what we need to do is go to everybody's house and make sure that only licensed statistical arbitrage traders have black boxes."

Worth reading for that line alone.

So we should invent our own uncorrelated black-boxen?

... that would analyze what "dumb" boxes are doing at the moment and do the opposite. If HFM is right about this, "dumb" boxes are moving big volumes of money actually.

Oh, that would be meta-black-boxes. A fine idea.

Perhaps more humbly - I was suggesting simple black-boxes that do behave differently. In statistical jargon: uncorrelated or even independent.

Everybody in the industry thought that their black-boxes were uncorrelated and independent. That was part of the appeal of them (and also why it's so difficult to sell algorithmic trading products - most firms want to develop them in-house). Problem was that they had no idea what everyone else was doing, and so most of them independently invented algorithms that behave fairly similarly.

It's like the thousands of entrepreneurs that saw Friendster and said "Oh look, only one competitor in the market, I'll build myself a social network and get rich!" Or the thousands of entrepreneurs that saw Reddit and Digg and said "Oh look, only two competitors in the market, I'll build myself a social bookmarking site and get rich!" Or the thousands of entrepreneurs that saw YouTube and said "Oh look, only one competitor in the market, I'll build myself a video sharing site and get rich!"

The only way to get rich is to build things where there are zero popular competitors in the market, or where there are 1-2 competitors and everyone believes it's fruitless to compete with them. Otherwise, you can bet that thousands of people you haven't heard of are doing the same thing, and by the time you get a product out they'll be getting a product out too.

(Incidentally, the first few quant hedge funds, like Renaissance or D.E. Shaw, made out like bandits.)

"Everybody in the industry thought that their black-boxes were uncorrelated and independent."

I'm not so sure. I interviewed at a Wall Street company that was programming data channels headed into the black-boxes. They boasted that 30% of trades were dealt with by computers. I mentioned that that should probably be enough to drive some pretty serious feedback effects. I asked: in principle, if these black-boxes handled all the trades, where would the decision to trade or not come from.

Instead of a blank stare, I was presented a cold, nervous look. He said "basically all these algorithms have been worked out in the 70's and 80's, in Academia. Nobody really knows what they're doing, or why they're doing it. But they are trying to compete by getting more data faster than the next guy. If everyone trades the same, as long as you trade faster, you'll win."

I asked if it was moral to play such a risky game with the economy.

I didn't get the job.

Everybody knows it in the sense that a debt collector knows that they're ruining people's lives, but it's kinda swept under the rug. We also had the experience of people saying "yeah, we don't really know how well our algorithms our doing, so if you can develop software to tell us, we'll pay you big bucks." (Though, last I heard, nobody had actually paid big bucks for the software we did develop.) But it's the kind of truth that nobody really wants to face, so they just kinda acknowledge it and then go about their business.

The only problem is that these funds don't start out trying to be correlated with each other--the most successful strategies are those that no one else is doing, since you, tautologically, get the best prices on broadly undervalued assets. They don't want to hold all the same things, but finance is a small world at people catch on to successful strategies pretty quickly. It just pans out that way since they are built on analyzing historical relationships between assets and what has worked in the past, e.g., buy refiners when the crack spread widens, or high p/e/g ratio stocks. To be independent you'd have to either find that relationship no one knows about, or pick a strategy that hasn't worked in the past.

To put it in web terms, it's almost like these quant funds are just adapting to industry best practices. We see that people like social networking, crowdsourcing, "web 2.0" page layouts, etc., so we see lots of sites racing to add these features--because hey, that's what works, that's what users want. But what happens if people get sick of one of those features (or any other you pick)? The relationship between the market and that feature breaks down. All those sites who counted on that strategy will all fail (or adapt) at around the same time, and many will rush into the next hot area (Pointcast-style "push" technology, anyone?).

Quant funds are basically just advanced machine-learners; you could implement a black box-of-sorts on your website by looking at Comscore numbers/trends for different sites and plotting that out against the features/layouts/topics they use, and instantly adding some new widget to your sidebar or something when you see a positive relationship with traffic generation. (Techmeme is a great example of a web black box, BTW, always on the hottest tech trend.)

I was at a financial software firm that was doing half of that. We were basically analyzing what the "dumb" black boxes were doing, but instead of trying to do the opposite, we just compared it with some well-known benchmarks so firms could see if their algorithms were actually doing well. At the moment, nobody actually knows what these black boxes are doing or whether they're actually performing better than the market as a whole.

While I was there, I thought about doing a product that'd try to predict the algorithms other people were using and profit off them, but you run into some really thorny epistemological problems. For starters, you generally can't find out which firms are making which trades, unless you're the firm itself (that's how our analysis product worked; it was for firms to evaluate their own performance, not their competitors). Most buy-side firms aggregate their orders through a few brokers, which keep their clientns' identities confidential. So you see "A sell order for 100 shares of MSFT at $33.01 came through Morgan Stanley", but you have no idea which fund placed the order or what other trades came from them. The signal gets lost in the statistical noise of all the other trades.

There's also no absolute standard of value in finance. Sure, you can benchmark values against discounted cash flow (assuming that you can even predict the cash flows, which is far from certain for most companies), but it may take years for a stock to revert to fair value as measured by discounted cash flow. Many of these automated trading systems operate over minutes or seconds, and don't care that the stock will eventually crash in 2-3 years.

one thing i don't understand is how can you make money by trading every couple minutes? because aren't there fees for every trade.

go read about subprimes here: http://unqualified-reservations.blogspot.com/2008/01/straigh...

Obviously fees are factored in.

Most of the folks who do this are the big financial firms who have a direct line to the exchange (actually, I've heard of boxes being colocated in the exchange to shave a few milliseconds off the trade time). They aren't paying retail commission rates.

There are still transaction costs though, mostly in the form of the bid-ask spread and the effect of moving the market when you buy or sell a lot of shares. Reducing these transaction costs is big business - one of the main areas that my employer was in was finding "hidden" liquidity where you can trade a lot without materially affecting the stock price.

Could you give an example of this hidden liquidity? Seems interesting.

It was something my boss did and I wasn't really involved in. As I understand it, though, it involved looking at NASDAQ Level 3 quotes (Level 3 basically gives you the full orderbook, so you have access to all limit orders placed on a certain security) and figuring out which stocks had the largest quantity for sale at a price near the current quote. It may also have had a statistical part to it too - that liquidity technically isn't hidden because it's visible for anyone who pays for L3 quotes, but you can then compare that to actual price movements after a certain quantity is traded, and then estimate what fraction of shares are visible in the order book as limit orders vs. what's sitting in computer algorithms ready to be placed as market orders if the price changes.

I heard a rumor that one guy who made money in August was Steve Cohen at SAC Capital -- he told his traders to run the model and do the opposite.

Wow. After reading that I realize just how little I know about finance.

I also now know how non-technical people feel when us programmers start talking in jargon.

It was an interesting article though.

HFM is a Slashdot reader apparently:

"HFM: Yes, but I for one welcome our computer trading masters."

More likely The Simpsons: http://en.wikipedia.org/wiki/The_Simpsons#Influences_on_lang...

awww bursting my tech loving bubble, of course this shows how much I know about television :)

Ick.

Reply on news.ycombinator.com