An agent based model of the E-Mini S&P 500 applied to flash crash analysis

Abstract

We propose a zero-intelligence agent-based model of the E-Mini S&P 500 futures market, which allows for a close examination of the market microstructure. Several classes of agents are characterized by their order speed and order placement within the limit order book. These agents' orders populate the simulated market in a way consistent with real world participation rates. By modeling separate trading classes the simulation is able to capture interactions between classes, which are essential to recreating market phenomenon. The simulated market is validated against empirically observed characteristics of price returns and volatility. We therefore conclude that our agent based simulation model can accurately capture the key characteristics of the nearest months E-Mini S&P 500 futures market. Additionally, to illustrate the applicability of the simulation, experiments were run, which confirm the leading hypothesis for the cause of the May 6 th 2010 Flash Crash.

Understanding flash crash contagion and systemic risk: A micro–macro agent-based approach

“…A consequence of that modelling decision is that their model does not generate stub quotes. As a result, the Paddrik et al (2012) flash crash price gradient appears shallow relative to an empirical flash crash or those generated by other models such as Vuorenmaa and Wang (2014). Our model relaxed the ten-tick assumption for the small trader agents only, permitting them to place orders at up to one-thousand ticks away from best.…”

“…Hanson (2011), Paddrik et al (2012), Vuorenmaa and Wang (2014), and Jacob Leal and Napoletano (2017) use combinations of agents representing low and high frequency market participants and are successful at reproducing the characteristic price fall and recovery observed during flash crashes. Indeed, the price recovery observed during flash crashes is another way in which crisis price impact differs from the linear models used in the fire-sales literature.…”

“…Detailed descriptions of the trader agent types are provided in Appendix A and are closely related to the implementations described in (Paddrik et al, 2012). However, the present model enhances the role of distressed selling agents from Paddrik et al (2012). The connection between the behaviour of the distressed selling agent and the liquidation behaviour of a controlling fund agent is discussed in the next section.…”

“…High Frequency Traders. High frequency traders (HFTs) in the model act as very fast market makers first and foremost, though they also possess a simple shared trading signal (Paddrik et al, 2012) which skews their order placement side probability in the direction of LOB imbalances. An imbalance in this context refers to a difference in the quantity of available shares resting on either side of the LOB.…”

Extracting information from the limit order book: New measures to evaluate equity data flow

“…Finance literature employs several avenues when studying HFTs activity. Agents-based simulation is one of the most widely used method (Leal, Napoletano, Roventini, & Fagiolo, 2016;Paddrik et al, 2012;Yang et al, 2012), which requires detailed trading data from each trader to simulate traders' behaviors. Similarly, Brogaard (2010) and Kirilenko et al (2017) rely on financial data sets containing specific trader information that allow to distinguish between HFTs and non-HFTs.…”