Finance at the speed of light: Is faster trading always better?
07/10/2014 | Marius Zoican – Vox
Technological advances in equity markets entered the spotlight following the Flash Crash of May 2010. This column analyses the advantages and disadvantages of algorithmic and high-frequency trading. Ever-faster exchanges do not always improve liquidity. Following a speed upgrade in the Nordic equity markets, effective spreads posted by high-frequency traders increased by 32%.
Few activities embraced the computer age so actively as trading. Loud and hectic pits have been progressively replaced by silent computer server rooms. Transactions are no less dynamic for it, however. A London-based trader can buy stocks in Frankfurt within just 2.21 milliseconds.1 Light needs 2.12 milliseconds to travel the same distance. Welcome to the age of algorithmic and high-frequency trading!
There is a very active ongoing debate around high-frequency trading. Supporters claim that in the few years since algorithmic trading took off, market liquidity and exchange competition improved. Critics point to aggressive high-frequency trading strategies that generate losses for human investors. Michael Lewis’ book Flash Boys (Lewis 2014) is a very well-known rendition of the latter view. How exactly does high-frequency trading affect markets? How can researchers and policymakers improve financial markets in the 21st century?
The benefits of algorithmic and fast trading
Why are high-frequency traders better at making markets? Their computer algorithms monitor in real time all information relevant to the traded asset: news headlines, demand and supply changes, or data on related assets. High-frequency traders are able to incorporate this wealth of information into their price quotes faster than anyone else. Two advantages follow directly. First, price discovery improves – price quotes accurately reflect all available information with minimal lag, as documented by Riordan and Storkenmaier (2012). Second, a savvy trader could have exploited the delay between news and price updates to earn a profit at the market maker’s expense. Fast trading minimises this delay, so the risk for a high-frequency market maker is lower than for a human one. Consequently, high-frequency traders are able to charge lower spreads. Trading costs are smaller for everybody (see Hendershott et al. 2011).
Algorithmic traders also promote competition between exchanges. In the past, assets were only traded at a single exchange. It made sense – having all potential buyers and sellers in one place increases the likelihood of finding counterparties. The exchange had a natural monopoly and the power to set large fees. Algorithmic trading made it easier to automatically search for counterparties across multiple exchanges (see, e.g., Domowitz and Benn 1999 and Menkveld and Yueshen 2013). Computer traders can take a position from a seller in one market and offload it to a buyer on a different one. There is no need for everybody to trade in the same place anymore. Under renewed competitive pressure, exchanges decrease trading fees.
The costs of algorithmic and fast trading
Are faster exchanges always good?
A starting point to answer this question is to acknowledge the empirical evidence showing high-frequency traders behave both as market makers and as speculators. Hagstromer and Norden (2013) document such order type specialisation. A speed improvement of a few microseconds directly affects high-frequency traders, irrespective of their strategy. Since human reaction time is hundreds of milliseconds, it does not directly affect human traders.
In a faster market, high-frequency market makers can update their quotes faster on new information. At the same time, high-frequency speculators are also faster to react to news. The market increasingly becomes a zero-sum game between these two types, with human traders being crowded out. Consequently, in low-latency markets, fast market makers are more likely to meet fast speculators. Whenever it happens, market makers are on the losing side; they are adversely selected. To compensate for the additional risk, market makers need to raise spreads.
We test the mechanism empirically, using a 2010 Nasdaq-OMX speed upgrade in three Nordic countries: Sweden, Denmark, and Finland. First, exchange latency dropped from 2.5 to 0.25 milliseconds. Second, traders were allowed to collocate with the exchange’s servers. Following the upgrade, the adverse-selection cost and the spread on quotes submitted by high-frequency market makers incr
A market design challenge