Facebook IPO: When Machines Collide
Transcript of Erik Townsend Interview with Nanex
What happens when machines collide in the most overhyped IPO of all time? Erik Townsend interviews Eric Hunsader of Nanex, giving an inside look at high frequency trading, quote-stuffing, and regulatory negligence. Excerpt provided with link to full transcript and audio interview below.
ERIK: Before we begin, I want to put Nanex into perspective for our listeners because to me the words “market data company” kind of imply if I want to know let’s say the closing price of IBM stock on a given day a couple of years ago, well, a market data company ought to be able to provide that information. But in the case of Nanex we're really talking about something much more detailed and different. For example, if I wanted to know three years ago on Wednesday afternoon March 17th, on a specific three-second period of time, let’s say between 2:37:14 and 2:37:17 in the afternoon, exactly precisely what happened with say IBM shares during that period, not only how many shares were bought and sold, but what prices were submitted, what limit order prices might have existed in the queue, Nanex can give me all of that information after the fact. So we're really talking about market data records that are so detailed that it becomes possible to do extremely in-depth analysis. It kind of reminds me of the analysis of a flight-data recorder after an airplane crashes or something.
Eric, give us a little bit more perspective on what you do at Nanex and what the company is about.
ERIC HUNSADER: We have about four trillion records now in our dataset and that comes from options, futures, equities, pretty much anything that trades in the United States. And not only can we tell you what happened in a three second window for a specific symbol, but we can tell you what other symbols and what other instruments were also doing at the same time.
ERIK: So you basically got all the data; everything that there is that is recordable, you have collected it, saved it and made it available for processing after the fact so you can analyze what happened.
ERIC HUNSADER: Well, actually, our business is selling this realtime to subscribers. That is our main business. The actual processing the data is more or less what we do for our customers when investigating oddities. [3:02]
ERIK: So you provide the realtime data feed and you also provide an add-on service which is the post-mortem analysis of interesting things that have gone on in the market.
ERIC HUNSADER: Correct.
ERIK: Now, before we dive into the gory details of what happened that fateful day with the Facebook IPO, I want to start by giving our listeners a sense of some of the background issues that came into play that day. Now years ago, a given stock say IBM stock was only traded on one and only one stock exchange. So IBM’s listed on the New York Stock Exchange, so that means the New York Stock Exchange is the only place that you could go to trade IBM shares, but that all changed a few years ago. Please explain to our listeners how and why it changed; what is Regulation NMS and how things work today.
ERIC HUNSADER: Regulation NMS is what was debated by the industry in 2005, 2006 and was finally implemented and rolled out in the first, second quarter of 2007. At the core of Reg NMS is this concept called the national best bid or offer, which each exchange trading a specific stock would submit their bids and offers; this information would be aggregated by what they called a SIP — there’s a lot of acronyms coming up here. SIP stands for Security Information Processor. And one of the formal names or actually implementation of the SIP you might hear is CQS, the Consolidated Quote System. That’s the same thing as the SIPs for a specific group of stocks. But anyways, so those SIPs would accumulate the bids and offers from all the exchanges, find the highest bid and the lowest offer and that would become the national best bid or offer so that an order from a customer coming to any exchange would have to trade at the best bid or offer before it could trade at a next lower price. That was called trade through price protection. And that is the core of what Reg NMS is all about.
ERIK: So essentially what Reg NMS does is it provides the industry with a way to make it possible for a certain stock or other security to trade on more than one exchange. And by having a set of rules that provide, first of all, an advertised national best bid or offer that assures that any buyer or seller can get a price that they know was the best price available and that their trades will execute at that price. So even though there might be multiple bids and asks on different exchanges, the Reg NMS system is intended to make sure everybody gets a fair deal, gets a fair price and all their trades execute at the best available price from any exchange anywhere in the world essentially.
ERIC HUNSADER: Correct. And by having a SIP that would collect the information from all the exchanges, an investor who wanted to receive and analyze prices on stocks wouldn't have to go to each exchange separately and contract for their data feeds because it would be all available in the SIP. And the SIP would also serve as an audit trail, so that the SEC or anybody could really go back in time and see exactly what happened at a given point in time.
ERIK: Okay. So the industry wants to do something fairly sophisticated, which is to trade the same securities on different exchanges. Regulators come back and say, okay, wait a minute, you guys want to do that, you've got to make it fair to investors by providing this thing called a SIP, which I believe stands for Securities Information Processor?
ERIC HUNSADER: That’s right.
ERIK: And by providing that, we're going to make sure that there’s one-stop shopping that any trader in the marketplace can go and look at what’s the best bid or offer that’s available on the SIP; all of the trades will occur at that price, everybody is protected and nothing fishy goes on behind the scenes in terms of how a given customer’s buy or sell order gets processed. Or at least that was the idea.
ERIC HUNSADER: Yes. It’s that simple. And it wasn’t the SEC mandating this on everybody, I mean this was — Reg NMS was invented by industry — widely respected people in the industry who made very good comments, some very prescient about it. So it wasn’t like rammed down everybody’s throats so to speak, it was after a pretty long public discussion period.
ERIK: Okay. Now there’s a concept in the regulation called the “eligibility for setting the national best bid and offer” and there’s also a concept of a “non-firm quote” that doesn't set that. Could you just explain that because that’s going to come into play in our analysis of Facebook in a couple of minutes here?
ERIC HUNSADER: That was pretty much to accommodate the NYSE which was still trading manually at the time versus NASDAQ the electronic exchange which was trying to say that NYSE might have a better bid than us, but you know, they’re doing it manually and they can’t execute it quite as fast as we can. We can execute it electronically. And so there was this concept introduced called “non-firm” where if an order was sent to the exchange and they didn’t respond within a certain amount of time, their quote could be considered non-firm and the exchange itself who was slow to respond was supposed to mark their quotes as non-firm, which told the SIP, hey, don’t include my bids and offers in your calculation of the best bid or offer.
ERIK: So essentially, a SIP’s job is to make sure that the best bid or offer being presented to investors is available in one place at any given time, and if an exchange is not able to participate in that system because its own quotes are screwed up for some reason or they can’t perform at an adequate speed or something like that, there’s a way for them to mark their quotes and just say, hey, don’t count us in this determination of the best bid and offer that investors are presented with.
ERIC HUNSADER: Correct.
ERIK: That all makes perfect sense as long as it’s being used in its intended mechanism, which, as we’ll see, doesn't always happen.
Now, another trend which has emerged in recent years that a lot of people have heard the buzz word but not everybody really understands is high frequency trading. Please give us a brief explanation of what high frequency trading is, who does it, how does it work.
ERIC HUNSADER: Well, you know, that term is kind of a catch all for many different strategies, all the way from stat arb and it’s just sped up to actual market making. We kind of define high frequency trading as trading algorithms that would not be profitable if not for speed. In other words, speed is what makes it profitable.
ERIK: So high frequency trading are computer programs instead of human traders that are trading the market and they have a variety of different algorithms that drive them. Some of them are just trying to scalp bid-ask spreads, others are trying to influence prices by buying and selling at certain times and so on and so forth.
ERIC HUNSADER: Right.
ERIK: What does the term “colocation” mean as it’s used with respect to high frequency trading?
ERIC HUNSADER: Well, the speed of trading has gotten to the point where the number one factor in how fast you can trade is the speed of light. And the speed of light travels in an idealized case approximately 186 miles in one millisecond. A thousand foot cable it takes the speed of light approximately a microsecond or a millionth of a second to travel down. And so the closer you get to the actual exchange execution engine the shorter your distance and the shorter your time and the faster you can trade. So colocation came about several years ago when the speed of trading got below say the 10 ms level.
ERIK: So if you’re sitting in your living room and you've written a high frequency trading algorithm and I’m sitting in my living room and I’ve written one, if one of us goes and spends a whole bunch of money to put our computer out of our living room and into the same building as the actual stock exchange where the network is connected not through the broader internet but through direct cables from my computer to the exchange, I’m going to have a substantial advantage over you trading from your living room because the messages are going back and forth much slower to your computer than they are to mine.
ERIC HUNSADER: Right.
ERIK: Proponents of high frequency trading have argued the presence of algorithmic trading is actually a benefit to all investors in the marketplace because these computer driven trades add liquidity resulting in tighter bid-ask spreads which benefit all investors. Please explain why you have a slightly different view.
ERIC HUNSADER: We can’t actually find any academic study that proves that spreads are tighter. In fact, just recently I believe it was the CEO of the NYSE said otherwise, that since Reg NMS the spreads have actually widened. And when you look at how rapidly the best bid and offer changes in one second period of time, it really is very difficult to measure what the spread exactly is. It’s like sometimes we’ll see the bid and offer change a thousand times in a second. It depends on which price point you decide to pick. So on the liquidity part, the liquidity, we've found, looks more illusionary than anything. In fact, one of the things that we saw from the Facebook IPO was when there was this glitch that caused an outage for about 17 seconds on all stocks on NASDAQ, as soon as that outage ended when NASDAQ came back online, we saw a tremendous amount of liquidity disappear from the books in just the blink of an eye in all stocks, not just Facebook and like SPY and other stocks. And it was a permanent thing for the rest of the day. In other words, as soon as these high frequency trading systems detect any problem in the system, they'll just stop trading. When I mean they'll just stop trading, I mean they'll stop trading in a fraction of a second. And so what you might see on the screen as liquidity is only going to be there as long as every thing is — if nothing has changed in that amount of time. But as soon as something goes wrong in the system or their algorithm doesn't like the input data they’re gone in a flash.