Financial firms that specialize in “high-frequency” trading (HFT) are now buying and selling stocks, futures, foreign currencies and many other assets with computer algorithms operating on timescales less than 1 millisecond. That's more than 1,000 trades a second. Modern markets have become complex ecologies in which millions of these “algos” compete all the time, and with advancing technology, 10,000 trades per second is just around the corner. More than half of all trading now takes place this way, and there's no end in sight to this “arms race” to ever faster speeds.
Non-financial people may rightly wonder: Why? Does it make sense? Is it fair, sensible, or possibly dangerous?
Indeed, it CAN BE dangerous, as such trading was directly implicated in the infamous Flash Crash of 6 May 2010, as well as in many “mini” flash crashes since then. Stock prices of course move erratically, but at least they used to move continuously; now stocks often jump discontinuously by several percent in a fraction of a second. Some experts, such as quantitative finance guru Paul Wilmott, have argued that high speed computer trading will bring on the next great financial disaster. Others, such as the Bank of England's Andy Haldane, point out that while high-frequency trading in ordinary times makes markets more liquid and therefore efficient, it also makes them more volatile and potentially explosive in times of stress.
I've written before about these risks associated with HFT, and remain convinced that we know very little about the things that might go wrong, especially as HFT becomes ever more dominant across different markets globally (see this worrying article, for example). Still, I'm not at all against technology when it can help us, if we can clearly understand how it does help us, collectively, rather than just the few parties who use it. So, I want to point out some far-from-obvious positive aspects of HFT, highlighted by recent research by physicist and finance expert Austin Gerig of the CABDyN Complexity Centre at Oxford University.
HFT firms don't go around telling people how they're trading. This secrecy has helped fuel controversy around what they do and also made it hard for researchers to tease out how exactly HFT activity affects markets. This is what Gerig has tried to do, using a special dataset supplied by NASDAQ, which reveals much greater detail on HFT trading activities. Perhaps the most striking thing he finds is that a primary effect of broad HFT activity is to “synchronize” security prices across financial markets. This means, for example, that if the price of Coke suddenly changes – because of fluctuations in sugar prices, perhaps – this will be followed almost instantaneously by similar price changes in other related securities such as Pepsi.
“Synchronization is a gargantuan task,” Gerig points out, given the huge number of stocks and other assets and the links between them, plus all the myriad derivatives products the values of which are directly tied to those assets. It's a task, he notes, that is “tailor-made for HFT as it is profitable for the firms that do it and can only be done with high-speed computerized trade.”
Ok, so HFT helps synchronization. So what? Using a standard model of financial markets, Gerig goes on to show that price synchronization is broadly good for the market, as it makes prices more accurate and thereby reduces transaction costs. Specifically, improved price accuracy leads to cost reduction because liquidity providers – market makers who stand ready to buy or sell at fixed prices at any moment – have more confidence that they won't be “picked off” or taken advantage of by someone out there who has better information. An important implication of this is that HFT is good for the average investor, as it makes real prices more apparent, and reduces the advantages of other investors with great information gathering and analyzing resources. It's the so-called “sophisticated” investors who actually lose out.
The intuition behind these results is straightforward. As an example, suppose that an event occurs which increases the likelihood that country X will default on its sovereign debt. This information is processed by specialized firms who quickly buy securities that track the probability of X's default. The prices of these securities increase, and if markets are synchronized, then the prices of all other securities adjust as well. As a result, an investor who purchases or sells any security in the market receives a more accurate price. Transaction costs are reduced because liquidity providers are more confident in market prices and require less of a price concession to transact with an order. In finance, this is known as a reduction in adverse selection costs.
If transaction costs are lower, then average investors benefit from synchronization. So, who loses? When prices are synchronized, information diffuses rapidly from security to security and informed investors are made somewhat redundant. In the model, they make less profit as a result.
So, it's not at all crazy that trades should take place at these time scales, as I've often thought it might be. Given the rate at which news now flows into markets, trading has to keep up if prices are to remain synchronized. It will – and probably should – get faster in the future.
Just one other thing, which is perhaps the most interesting. Gerig points out that this synchronization effect — here between the prices of different assets — is actually a direct analog of the way flocks of birds or schools of fish synchronize their movements. That effect is know to provide a highly efficient way for a population to harvest information from “many eyes” (see this nice essay at Nature, for example; subscription necessary) to make important decisions – such as fleeing a predator – both more accurately and faster. Watch the amazing video below of a school of X fish evading sharks, as well as a hungry seagull, with remarkable speed and precision.