"Information Content of Book and Trade Order Flow at Different Time Scales" (joint work with Alessio Sancetta and Yuri Taranenko) [under Review @ Quantitative Finance]
Abstract:
This paper studies information spillovers from the NASDAQ Limit Order Book (LOB) and assesses their impact on price predictability. Using LOB data for 35 large cap US stocks from March 2019 to February 2023, we aggregate data at different trading volume time scales and train various machine learning algorithms: Linear Discriminant Analysis, Ridge Classifiers, Random Forests and a Deep Neural Network. Our empirical findings suggest that trade order flow information is the most persistent and prices are predictable with respect to it. We document that machine learning models are able to predict mid price directions accurately, yet this informational advantage dissipates within the first 10 milliseconds. Moreover, our findings suggest that model complexity does not necessarily ensure higher financial returns. Using information on quoting activity from other exchanges, we also conclude that market participants may choose to quote more heavily on the NASDAQ, but they do so without leaking more information. Additionally, employing both panel and cross sectional analysis, we examine how stock-specific and market determinants affect intra-day predictability across different days. Overall, more liquid stocks with higher market beta exhibit higher intraday returns. We show that there is persistency in high frequency performance and the dynamic adjustment towards the long-run average lasts up to three trading days. Finally, we document that over time, the value of order flow has decreased. Then, it is plausible to infer that the growing use of algorithmic trading has increased market competition and consequently enhanced market efficiency.
"MINT: Multi-participant Interactive Trading for Experimental Economics" (joint work with Wenbin Wu)
Abstract:
We present the Multi-participant Interactive Trading (MINT) platform, an experimental enviroment for studying trading behaviour in continuous double-auction markets. MINT supports both human-human and human-machine trading interactions, enabling experimental treatments involving algorithmic traders. The platform is buit using Python and provides transparency and customisation through an open-source architecture. Researchers can integrate custom trading algorithms and modify experimental parameters through an administrative interface without platform redeployment. MINT records Level 3 (L3) quote data, including all order submissions and message logs, facilitating analyses of market and order book dynamics and individual participant behaviour. The platform integrates with participant recruitment services such as Prolific. Future developments include LLM-based decision systems to explore AI-assisted trading behaviour.
"Bayesian Inference in Dynamic Panel Stochastic Frontier Models" (joint work with Mike Tsionas and Marwan Izzeldin) [RR @ JRSS: Series A]
Abstract:
The paper adopts a dynamic panel stochastic frontier model that incorporates firms' intertemporal decision behaviour and short-run stagnant adjustments to the production process. Its dynamic specification recognises short-run output adjustment costs, where final output may be only partially adjusted to the optimum level. In nesting previous panel stochastic frontier models, our new approach delivers a flexible framework that accommodates heterogeneous technologies and latent time-varying inefficiency effects. Model inference is based on Bayesian Markov Chain Monte Carlo (MCMC) techniques with data augmentation. Using simulated data, we illustrate that our model performs very well in small and moderate samples. Last, we present our model in an empirical example, analysing all publicly listed UK companies operating in the manufacturing and construction sector over the period 2004-2022. A general finding is that most firms exhibit stagnant production processes, with the half-life for adjusting supply to be as high as 5-quarters. The estimated technical efficiencies range from 43% to 96% across the years.
"Human-Machine Inreraction in Electronic Financial Makrets: Experiments" (joint work with Francensco Feri, Alessio Sancetta, Michael Naef and Wenbin Wu)
"Endogenous Technical Efficiency: An application to UK manufacturing sector"