Trading Shares in Milliseconds | Hacker News Reader

Trading Shares in Milliseconds

technologyreview.com

64 pointsraju30 comments

All the major equity exchanges have a signal on their feed which indicates "all stop" to automated engines. It's a signal which automatic trading systems are required to listen to and on receipt suspend trading. Historically it's activated by the exchange if volatility has gone too high or if the market is behaving abnormally. It's one of the many safety systems the financial system has in place to protect against an algorithmic meltdown.

Meltdown being the apt term, as the safety precautions aren't dissimilar to what you have in a nuclear system. You have multiple levels of checks-and-balances. For something to go seriously wrong 11-12 things have to break in combination. Obviously this isn't impossible and can (and probably will) happen sooner or later, but it in everyone's individual interest as well as the collective interest to prevent a collapse.

The big problem isn't the risk that an investment firm will screw up and do bad deals (in practice bad share deals due to system error tend to be reversed by mutual agreement by the counterparties; banks know systems screw up and generally are willing to reverse such deals), but rather some new form of systematic risk is created which isn't properly protected against. Dealing with systematic risk is precisely where the regulators should be coming in because it's outside the capability of any individual institution to deal with it.

> Dealing with systematic risk is precisely where the regulators should be coming in because it's outside the capability of any individual institution to deal with it.

Actually, regulators are quite good at creating systemic risk.

The widespread ownership of Fannie and Freddie stock by banks, which killed their balance sheets when Fannie and Freddie went down, came from regulation. So did the popularity of securitized mortgages and "insurance". (Regulators required insurance, AIG's fit the bill, and everything was great, until it wasn.t)

And, let's not forget that Wells Fargo got hammered because it didn't go along with the regulators' "requests" to do dumb home loans.

Many markets have a notion of the a volatility break. If the price moves too far in a short period of time, trading on the stock is suspended for a few minutes, while limit orders are collected. An auction crossing is then held to reopen the stock. This can happen several times in a row.

Could the problem possibly be solved in a different way, for example a different kind of market or with different transactions (or by publishing information differently)? I have a feeling that the existing stock exchanges deliberately leave the system as it is, because they earn money from the high frequency traders.

As an example for what I mean: if there was only one stock exchange, there would be no arbitrage between different stock exchanges. Or maybe transactions could come with a delay. Or buyers and sellers could be more open about their prices. Essentially, what would the market look like if the high frequency traders were 100% efficient - couldn't it be constructed to be like that without the high frequency trader's involvement?

The one example where the high frequency traders can determine the exact price by ordering and canceling within milliseconds certainly does not sound very fair.

There are lots of stock exchanges which offer alternative algorithms for order-matching. It's up to each individual company which primary exchange it chooses to be listed on (although obviously it can't stop interested parties creating a secondary exchange using whatever rules they want between them).

Some exchanges specifically prohibit the practice of flashing while others don't. It's a free-market, if people and companies want to trade on exchanges with alternative matching algorithms and prohibitions on flashing, there's nothing stopping them from doing it.

Exchanges are only as valuable as the people who trade on them, if the traders want another set of rules than the exchanges have to adapt otherwise the traders will go elsewhere.

Sure - so I guess if the hf traders really were a problem, people would start migrating to other platforms.

The conclusion of the article: Automated Trading could be dangerous to the market as such and should be closely monitored. They appear disturbingly similar to the hedge funds of the old days. Besides, quant-finance big shots like consider automated trading based on algorithms risky.

Correct me if I'm wrong.

Actually, I think this argument (posed by the journalist) was woolly. Given that the high speed trading firms seem to stay market neutral (each trade is adequately hedged) and unwind their trades within hours or days, their inherent risk is less than that of massive pension funds withdrawing their capital as we saw in the 07/08 crash.

As was said in the article, these guys are really just providing liquidity for smaller traders (i.e. you and me) who want to be able to move on a security in small chunks without having to pay for an expensive broker.

That said, there is a crash-scenario if the traditional (i.e. non high-speed) platforms are triggered via stop loss mechanisms to offload lots of stock fast. But again, this shouldn't happen if they do two things:

First: look at the source of the trade. If it's one of the high speed firms, they should just ignore the trade unless it moves with a velocity of say 25% more than the high speed stuff - i.e. if it's dropping by 1.25% of the price per minute rather than 1%, then it should consider selling. (or, for a more general formula - sell if total_velocity - highspeed_velocity = x/s)

Secondly, take wider market movements into consideration - if the entire sector is falling, it's probably not an issue with the security, so there's really no point in getting out at that point - it should recover.*

The one technique revealed in that article which i found interesting - pinging to find the edges of the applicable range trade- guessing the probable worst case purchase by offering and withdrawing stock to see if it's jumped on. I could imagine this getting banned, along with the 'who's trading soon' information.

* of course, this depends on the time period the stock is willing to be held for - if the goal is day trading, then you're not going to want to stick around.

I work in this area, and the market risk you allude to here is not the one that scares me. The scary one is what I like to call algorithm or system risk. The 2003 Corinthian Colleges incident mentioned on page 3 of the article is a good example, but I've seen much worse. I once saw a major bank take literally a billion dollars of risk on in approximately 10 seconds due a bug in their algorithm (most likely a new release of a modified program) that interacted with a poorly implemented exchange interface. Some of their counterparties let them out of the trades and some didn't. Releasing new software can be a hair raising experience due to the poor quality of testing systems provided by the exchanges, the speed at which money losing trades can happen, and the vast complexity of the environment faced by the software.

I wonder how long / how hard it'd take to implement a shadow exchange to test new algos and systems... Like, perhaps new code needs to be certified in a full test environment before being deployed to the live market?

Seems like something along those lines would adequately deal with the uncertainty risk here.

I think one of the problems comes from the fact that big players affect the game. So if say, Citi launches a new algorithm that does some new crazy thing, which turns out to trigger random issues or unexpected responses in other algorithms -- such a thing could make lots of people other than citi a pile of money. Are those people really going to want Citi to test it too well?

Another, real concern with such a shadow market, where say everyone is running their algorithms (to prevent the above scenario), is that i could test "a buggy version" and find holes and odd behaviors of my competetors for free, since it is all play money.

Most exchanges have their own API for direct access. Although there is some standardization around the FIX protocol, the fastest access is still usually through the direct API, so while you may have a single backend system that processes incoming data and generates orders, you still usually need to code a separate interface for each exchange that you connect to, and the systems provided by the exchanges to help you test your interface tend to be less than robust.

Your last two sentences are a bit vague. When are the "old days"? 30 or 40 years ago, or pre 2007/8 financial crisis? Automated trading is risky; but then all trading is risky. Do you mean as opposed to another type of trading?

@andrew1, In the first sentence I referred to the 2000-2008 hedge fund trading. The second sentence is from the article - it said high speed algorithmic trading could lead to smaller version of the 1987 crash.

Oh I see, my apologies, I thought that the first sentence was the summary, and the second two were your thoughts on it.

From what I remember from Greenspan's "Age of Turbulence," he mentioned that he was pleasantly surprised that the markets were able to absorb much of the shock that happened on 9/11. I believe he attributed much of that to flexibility in the markets, which is probably related to the liquidity and efficiency that are provided by, for example, high-volume traders. Greenspan argues for deregulation of such activities since they make the markets so much more flexible, and able to withstand sudden shocks such as 9/11.

Of course, much of that was before the fall '08 crisis, so I don't know how he'd respond to that :)

http://www.businessinsider.com/high-frequency-trading-hits-t...

The Daily Show does an amazing job with this subject.

How did they do an amazing job at all? The linked video (picture #7 is a video) was far more productive.

I always wonder what general sort of "algorithms" they use. Neural nets?

This is what most of my research has been. Here's a post from Markenomics that I created some time ago with a lot of links on the algorithms used.

http://www.markenomics.com/item?id=321

Also, the US Patent website has lots of interesting stuff if you are really wanting to dig deep into this stuff.

That site has tons of information. I wonder how come I've never heard of it. Thanks a ton.

Here's an example of what the demigods at RenTec do:

http://www.bloomberg.com/apps/news?pid=newsarchive&sid=a...

"Volfbeyn said that he was instructed by his superiors to devise a way to 'defraud investors trading through the Portfolio System for Institutional Trading, or POSIT,' an electronic order-matching system operated by Investment Technology Group Inc. Volfbeyn said that he was asked to create an algorithm, or set of computer instructions, to 'reveal information that POSIT intended to keep confidential.'"

Now, you didn't think they used Black-Scholes, did you?! If you did, then: welcome to the real world!

for instance: http://en.wikipedia.org/wiki/Technical_analysis

I am wondering what clueless fools have downvoted you. Technical Analysis is as foolish as Black-Scholes. Maybe moving-averages and Bollinger bands are not algorithms for the HN crowd. Technical Analysis is retarded, but then... so is all that crap they teach at Quantitative Finance courses.

Er, what? Do you have any idea what you're talking about?

Actually, I do. Do you?

I found it strange that a guy got downvoted for suggesting that a hedge fund used technical analysis. I hate TA, but then, I hate Quant Finance, too. Whoever thinks that the smart guys at RenTech and the like use that kiddie Stochastic Calculus taught at MFE programs is living in a state of sin. Period.

Technical analysis is just voodoo. I don't know what you mean by "quant finance" but stat arb is merely played out.

The finance academics borrowed the mathematics from physics and the rigor from mathematics, but they forgot that, at the core, theirs is a social science. If they wanted to borrow anything from physics, they could have borrowed the good habit of comparing their theoretical results with experimental data. In physics, a theory is little more than mental masturbation until it's confronted with experiment. That nicely summarizes the state of contemporary academic finance: mental masturbation. Just because physicists were blessed with the ability to model natural phenomena using mathematics, it does not mean other scientists have such luck.

Have you actually worked in this field? As a quant?

I did. I was unimpressed. I was used to working in fields where theories could be trusted. Finance is a lawless territory. It's fun because problems lack structure and are so hard. It's frustrating because 30 years ago my skills would have allowed me to do something cool, while now the field is saturated and the odds are stacked against you. I moved to greener pastures and never looked back.

Reply on news.ycombinator.com