Since the inception of the Amsterdam stock exchange in 1602, the world’s oldest stock exchange, more than 100 exchanges have been established around the world. If Apple stock were trading in every exchange, you can be sure that even our fastest trading systems wouldn’t be able to tell you to which exchange you should send your order flow in order to match the best quote using the best currency exchange.
Were we ever able to do so? In the past year, it was fun to watch how bitcoins were traded in more than 10 unregulated exchanges around the globe. A great example of how arbitrage could, and does work. But putting regulations and the various types of securities aside, no trader could have access to all the quotes from all the exchanges simultaneously as they become available in any market. The best solution, thus far, was to deploy High Frequency Trading (HFT) systems as far back as a decade ago in hedge fund companies such as Renaissance Technologies and Algorates.
An HFT system is an artificially intelligent trading system which deploys computer algorithms (“robots”) to acquire and process a high volume of quotes quickly and assist their respective brokers in making trades. All portfolio-allocation decisions are made by computerized quantitative models. Robots move in and out of short-term positions on non-random, easily traded stocks to secure low margins (less than a cent in profit per trade) using incredibly high trade volumes, typically in the millions of shares. HFT firms don’t hold positions overnight. Too risky! But one thing is clear: with the exception of hackers, no HFT system can see quotes before they get to the market. Once they are in the market, however, HFT systems can grab them first, literally split seconds ahead of the low frequency trading system.
HFT systems rely on the speed of fiber optic cable to send/receive data. There is a latency due to the speed of light (approximately 1 foot per nanosecond), thereby limiting the data flow to one billion quotes per second. Depending on where a trading system is located and how fast the network is, they may not be able to see all quotes available in all the markets. As a result, the trading system won’t get the best possible bid/offer price, whereas an HFT system located physically closer to exchanges and having access to a bigger pool, finds a better price for the investors, or may even manipulate the market. Don’t blame all HFT systems; judge the strategy used by robots in the HFT systems that upset the financial systems. For heavily traded stocks (e.g. having 20,000 or more quotes per second multiplied by 2,000 active securities), an exchange could potentially get 20 million quotes a second. What if an HFT system sends 20 billion trades per second instead?
NBBO (National Best Bid or Offer) is an SEC rule that is designed to protect investors and traders to ensure that trades are routed to the market offering them the best bid/ask price. As an example, if NASDAQ is posting a bid of 100 shares at $500 and an offer of 500 shares at $510 in security APPL. and another exchange is posting a bid of 200 shares at $490 and an offer of 200 shares at $507, the NBBO would be 100 shares at a $500 bid and 200 shares at a $507 offer. Very simple and straightforward with an assumption that all trading systems have access to all these quotes instantly. However they don’t. And there is no minimum quote-life requirement in the SEC rule. The fact is Security Information Processor (SIP) feed is comparatively slow and an HFT system can send and cancel a quote in less than a microsecond.
This latency is one of the reasons nobody really uses the SIP for NBBO anymore and each exchange calculates NBBO internally in a computer’s RAM (while connecting to other exchanges with no audit trail). As the speed of trading increases, the gap between different exchanges’ NBBO increases. The beloved SEC rule appears to be less enforced by exchanges and hedge funds as the quote lifetime approaches zero.
It should come as no surprise that NYSE floor traders recently spoke in less than enthusiastic tones about HFT. HFT accounts for over 60 percent of the trading volume coming from 14 U.S. exchanges over the past 5 years. And the number keeps going up, with the implication that HFT systems will take away more jobs from traders. Predictably, traders are not happy with this movement in exactly the same way auto dealers are not happy with Tesla. But the reality is that money manager institutions will eventually have to move to HFT systems to protect their clients or they will lose those clients.
So what’s wrong with receiving faster responses from all exchanges? That’s what HFT promises and delivers perfectly in addition to lowering operational risk and transaction fees. The issue is which algorithm (strategy) HFT systems use to execute trades once they know the market condition. Do they act in the best interest of their investors? Do they keep records of all quotes from other parties when communicating with different exchanges? Low frequency trading systems or traditional systems could have the same issues as well. Investors can lose opportunities because the system is too slow combined with the impact of human error. Sit down with an honest trader and they will tell you how many times they’ve screwed up. There are bad apples in either system. One is going to retire and the other one is not perfect but it’s getting better over time.
But HFT’s concept is not just about speed. It comprises strategies that could potentially provide liquidity to the markets. It’s interesting to see how strategies work under pressure when markets tank or boom. How do you know which HFT system you should use as a money manager? If an HFT system is faster due to its proximity to the exchanges and faster network and processor, but without the best strategy, who will lose? Are their strategies NBBO certified? I doubt if SEC could force any NBBO rule without introducing a faster consolidated SIP-like real time system.
Hedge funds and money manager institutions are the major high frequency players. They use different strategies for micro and macro events. These strategies have different holding periods from microseconds to a full day and execute quantitative algorithms for optimal pricing, trading on events, trading on arbitrage, just to name a few. The HFT strategies are more complicated than the traditional buy-and-hold strategy. But their goals are pretty much the same: maximize ROI and minimize the overall risk. As a math whiz/computer scientist working in an HFT system to build the most profitable strategy, you would be looking into how to factor time and access to better quotes in your strategies to benefit the traders. You could test your strategies in a fully automated trading system as compared to traditional systems. You can back test your strategies with historical data. You can stress test them with volume going to any direction. Basically you have a better way to implement more strategies and measure them much more quickly. I am not saying that every HFT system provides users with all these features. I am suggesting to get one that has them.
I believe SEC will soon enforce a minimum quote life to all HFT systems which could change a lot of existing strategies. As more trading systems move to HFT, combined with a minimum quote life policy (let’s say 100 microseconds or so), HFT performance will be very similar among different systems in the future. Having said that, their survival will rely heavily on a firm’s budget. HFT systems will need to constantly upgrade their strategies, hardware and software, not unlike the mining of bitcoins.
