Roy Hayes

An Agent Based Model of the E-Mini S&P 500 and the Flash Crash

We propose a near zero-intelligence agent-based model of the E-Mini S&P 500 futures market that allows for a close examination of market microstructure in the context of a flash crash. Several classes of agents are characterized by how fast they trade and where they place trades in the limit order book. These agents’ orders populate the simulated market in a way consistent with real world participation rates. The simulated market is validated against important empirically observed characteristics of price returns and volatility. Additionally, to illustrate the applicability of the simulation experimental results are present, which examine the leading hypothesis for the cause of the May 6th 2010 Flash Crash.

Agent based model of the e-mini future: application for policy making

An agent-based model (ABM) has a structure, which includes a set of agents, a topology and an environment. A simplified conception of a financial market includes a set of market participants, a trading mechanism, and a set of securities. In a typical ABM of a financial market, the market participants are agents, the market mechanism is the topology and the exogenous flow of information into the market is the environment. A zero-intelligence ABM model of the E-Mini Futures Market is presented. Several classes of agents are characterized by their speed and placement of orders within the limit order book. The proposed minimum quote life rule is implemented in the simulation. The minimum quote life rule prevents new orders from being cancelled or modified before a given time limit. Through experimentation, tradeoff curves are generated. Thereby, illustrating the usefulness of this ABM and its ability to inform ongoing financial policy debates.

Glucose Control, Diabetes Status, and Mortality in Critically Ill Patients: The Continuum From Intensive Care Unit Admission to Hospital Discharge

Objective: To describe the relationships among glycemic control, diabetes mellitus (DM) status, and mortality in critically ill patients from intensive care unit (ICU) admission to hospital discharge. Patients and Methods: This is a retrospective investigation of 6387 ICU patients with 5 or more blood glucose (BG) tests and 4462 ICU survivors admitted to 2 academic medical centers from July 1, 2010, through December 31, 2014. We studied the relationships among mean BG level, hypoglycemia (BG level <70 mg/dL [to convert to mmol/L, multiply by 0.0555]), high glucose variability (coefficient of variation ≥20%), DM status, and mortality. Results: The ICU mortality for patients without DM with ICU mean BG levels of 80 to less than 110, 110 to less than 140, 140 to less than 180, and at least 180 mg/dL was 4.50%, 7.30%, 12.16%, and 32.82%, respectively. Floor mortality for patients without DM with these BG ranges was 2.74%, 2.64%, 7.88%, and 5.66%, respectively. The ICU and floor mean BG levels of 80 to less than 110 and 110 to less than 140 mg/dL were independently associated with reduced ICU and floor mortality compared with mean BG levels of 140 to less than 180 mg/dL in patients without DM (odds ratio [OR] [95% CI]: 0.43 (0.28–0.66), 0.62 (0.45–0.85), 0.41 (0.23–0.75), and 0.40 (0.25–0.63), respectively) but not in patients with DM. Both ICU and floor hypoglycemia and increased glucose variability were strongly associated with ICU and floor mortality in patients without DM, and less so in those with DM. The independent association of dysglycemia occurring in either setting with mortality was cumulative in patients without DM. Conclusion: These findings support the importance of glucose control across the entire trajectory of hospitalization in critically ill patients and suggest that the BG target of 140 to less than 180 mg/dL is not appropriate for patients without DM. The optimal BG target for patients with DM remains uncertain.

Action-based feature representation for reverse engineering trading strategies

This paper considers the problem of reverse engineering strategies for trading in the financial markets. We investigate this problem in the context of a trading tournament in which student teams used delta hedging and other mechanisms to attempt to achieve benchmark performance in managing a hedge fund in a simulated market. Our hypothesis is that machine learning models can be trained to solve the apprenticeship learning problem; that is, these models can learn to trade like tournament participants. After reviewing classical return-matching approaches and recent work in inverse reinforcement learning, we propose a supervised learning methodology that makes use of recursive partitioning (RP). Our proposed RP approach is based on a feature representation for actions that, we argue, corresponds to the information structures readily available to tournament participants. RP achieves high accuracy in predicting the type and scale of participant trades and in tracking overall portfolio performance. Our results suggest that further research on our proposed approach is warranted and should include an expansion to testing on data from real markets.

Agent Based Model of the E-Mini S&P 500 Future: Application for Policy Making

An agent-based model (ABM) has a structure, which includes a set of agents, a topology and an environment. A simplified conception of a financial market includes a set of market participants, a trading mechanism, and a set of securities. In a typical ABM of a financial market, the market participants are agents, the market mechanism is the topology and the exogenous flow of information into the market is the environment. A zero-intelligence ABM model of the E-Mini Futures Market is presented. Several classes of agents are characterized by their speed and placement of orders within the limit order book. The proposed minimum quote life rule is implemented in the simulation. The minimum quote life rule prevents new orders from being cancelled or modified before a given time limit. Through experimentation, trade-off curves are generated. Thereby, illustrating the usefulness of this ABM and its ability to inform ongoing financial policy debates.

Computational intelligence in financial engineering trading competition: a system for project-based learning

This paper discusses the implementation of the Trading Competition held at the 2014 IEEE Computational Intelligence in Financial Engineering conference (CIFEr 2014). Participants in the competition were asked to hedge a simulated portfolio of assets, worth approximately

An agent-based financial simulation for use by researchers

54 million. The winner was the individual whose portfolio most closely generated a 1% annualized return based on daily tracking. The goal of the competition was to provide participants with the opportunity to learn portfolio management and hedging skill. Self-assessments indicate that contestants improved their portfolio management skills and enjoyed their experience. This paper discusses methods used to generate the simulated stock and option prices and to construct the trading platform. All of the software used in the competition is being made open source in the hope that students, professors, and practitioners improve on the idea of the competition, thereby facilitating project-based learning for the future practitioners of economics, finance, and financial engineering.

Revolutionizing Financial Engineering Education: Simulation-Based Strategies for Learning

Regulators and policy makers, facing a complicated, fast-paced and quickly evolving marketplace, require new tools and decision aides to inform policy. Agent-based models, which are capable of capturing the organization of exchanges, intricacies of market mechanisms, and the heterogeneity of market participants, offer a powerful method for understanding the financial marketplace. To this end, we have worked to develop a flexible and adaptable agent-based model of financial markets that can be extended and applied to interesting policy questions. This paper presents the implementation of this model. In addition, it provides a small case study that demonstrates the possible uses of the model. The source code of the simulation has also been released and is available for use.

Unsupervised Hierarchical Clustering of Build Orders in a Real-Time Strategy Game

On May 6, 2010, world financial markets experienced The Flash Crash, which affected trillions of dollars of securities in just mere minutes. Consensus of what caused this dramatic change in the market evaluation of thousands of assets is still not fully understood. To tackle problems like these, financial engineering students need a comprehensive understanding of complex market microsystems. These students must have a multidisciplinary skillset that incorporates an ever- widening array of disciplines to address markets that have grown increasingly tangled. Fundamental understanding of today’s financial markets requires students take courses in statistics, mathematics, computer programing, finance, and economics; however, due to the limited material that a single course can cover, traditional coursework cannot effectively teach the multidisciplinary competencies that are necessary.In order to reduce the course load yet still learn fundamental financial engineering principles, we suggest taking a constructivist approach using market simulations as teaching tools. Simulations allow financial engineering students to learn the complex nature of markets more comprehensively and effectively by allowing them real-world experience in controlled, guided environments. Research supports their use and has shown simulations engage students and give them real-time knowledge of cause and effect in a complex marketplace. Moreover, they go beyond simply providing factual knowledge to offering students experience operating in the microstructure of markets. This environment also forces students to encounter unanticipated situations that traditional education methods do not allow.In that spirit, this paper will present the use of simulations in a simulation-based lesson that will teach students how to devise and implement their own trading strategies, as well as give them invaluable experience in integrating the basics of statistics, programing, and design. This paper adds to the body of research that illustrates simulations have the ability to get students emotionally invested in learning and thereby more receptive to the minutia of financial markets and trading techniques. Simulation-based education can augment traditional education methods by providing a learning environment that promotes more skills and techniques to foster better fundamental knowledge.

The need for a real time strategy game language

Currently, no artificial intelligence (AI) agent can beat a professional real-time strategy game player. Lack of effective opponent modeling limits an AI agent’s ability to adapt to new opponents or strategies. Opponent models provide an understanding of the opponent’s strategy and potential future actions. To date, opponent models have relied on handcrafted features and expert-defined strategies, which restricts AI agent opponent models to previously known and easily understood strategies. In this paper, we propose size-first hierarchic clustering to cluster players that employ similar strategies in a real-time strategy (RTS) game. We employ an unsupervised hierarchal clustering algorithm to cluster game build orders into strategy groups. To eliminate small outlying clusters, the hierarchal clustering algorithm was modified to first group the smallest cluster with its closest neighbor, i.e., size-first hierarchal clustering. In our analysis, we employ a previously developed dataset based on StarCraft: Brood War game replays. In our proposed approach, principal component analysis (PCA) is used to visualize player clusters, and the obtained PCA graphs show that the clusters are qualitatively distinct. We also demonstrate that a game’s outcome is marginally affected by both players’ clusters. In addition, we show that the opponent’s faction can be determined based on a player’s transition between clusters overtime. The novelty of our analysis is the lack of expert-defined features and an automated stopping condition to determine the appropriate number of clusters. Thus, the proposed approach is bias-free and applicable to any StarCraft-like RTS game.