Experiments in high-frequency trading: comparing two market institutions

Abstract

Aldrich and Friedman (2017) obtain proprietary data from the IEX exchange for the month of December, 2016. IEX classifies each participant as either an "agency" or "proprietary" trader, the former being the class of traders with a fundamental interest to buy or sell assets (i.e. to maintain an inventory for portfolio reasons). Aldrich and Friedman (2017) report that IEX agency transactions comprised 10,498,518 shares of the S&P 500 exchange traded fund (ticker SPY) during the 21 trading days or 21 × 6.5 × 60 = 8190 trading minutes during December, 2016. Since the median trade size is the minimum block of 100 shares, this amounts to 10, 498, 518/100 ≈ 105, 000 total trades during the month, or 105, 000/8190 = 12.82 investor arrivals per minute, or roughly 1 investor arrival every 4.68 seconds. Although IEX represents only a small fraction of equities market share, we believe that most fundamental traders (investors) will utilize IEX in conjunction with other equities exchanges, and hence that their arrival rates would be suggestive of aggregate investor arrival intensities. The result is that in raw time (we discuss time rescaling below), the Poisson intensity parameter for investor arrivals isλ I = 1/4.68.

Do speed bumps curb low-latency trading? Evidence from a laboratory market

“…After each trader chooses their latency investment level, we simulate a continuous-time market where orders execute at random times against a simulated market maker. In some sense, this setup echoes the mechanics of algorithmic trading (as in Aldrich and López Vargas, 2019), as traders first define strategies which are subsequently implemented by computer software.…”

Market Design, Human Behavior, and Management

“…Moreover, although there are three decades of studying financial markets in the laboratory (for surveys on experimental research in financial markets, see Friedman and Rust 1993, Friedman 2010, and Noussair and Tucker 2013, aside from particular episodes such as the Flash Crash (Aldrich et al 2016), little is known about the impact of sniping in times of financial stress as opposed to normal times (but see Jagannathan 2019 for a step in this direction). However, Aldrich and López Vargas (2019) recently conducted a laboratory market design study on high-frequency trading that suggests that, relative to the continuous double auction, the frequent batch auction exhibits less predatory trading behavior, lower investments in low-latency communication technology, lower transaction costs, and lower volatility in market spreads and liquidity. More studies on how financial market design affects sniping, market stability, and market resiliency are necessary.…”

Algorithmic Trading in Experimental Markets with Human Traders: A Literature Survey

“…In a more complex CDA market design with a multi-period lived asset, which frequently generates bubbles and crashes in laboratory studies (i.e., the design of Smith et al 1988), 3 Duffy and Ünver (2006) also report price and volume paths similarities of their enhanced "near" ZI-trader markets with the human-trader experimental results of Smith et al (1988). In particular, the authors are able to reproduce bubble and crash patterns in the simulated "near ZI" market.…”

“…Different from the ZI of Gode and Sunder (1993), the "near ZI" trader's bids and offers are not purely random but are biased towards the past period's average price. 3 The design of Smith et al (1988) is described as follows: Nine subjects, initially endowed with cash and assets, can buy or sell assets between each other during T=15 periods in a CDA market. Note that differently from the Smith (1962) market there no induced values and costs, but common values, and each trader can buy and sell assets.…”