Commodity trading is one of the most ancient market activities. However, compared to other popular financial instruments, the trading of commodities is probably the only few that still resist automation. One of the major reasons why automating commodity trading is difficult is the physical implications behind the act of trade. When a trade is completed, physical movement of goods must actually take place, and a great number of physical issues need to be considered besides just profit and loss in trading. This unique consideration is reflected in the fact that a wide variety of physical events would have indirect or direct impact on the dynamics of the commodity market. Added to the complexity is the fact that different commodities might have very different set of issues/events that would affect their market dynamics. As a result, it is extremely challenging to develop a generic analysis framework for studying commodities.

In this research, we propose to adopt a simulation approach in studying commodities. With a simulation system that sufficiently describes the important features of a commodity market, we can then make progress on automating the management of a commodity supply chain. The foundation of our simulation is multi-agent-based, and combines both event-based approach and event study method for market dynamics generation and validation. In our event-based approach, the simulation is progressed by announcing news events that affect various aspects of the commodity supply chain. Upon receiving these events, market agents that play different roles, e.g., producers, consumers, and speculators, would adjust their views on the market and act accordingly. Their actions would be based on their roles, their private information, and also the event information, and collectively the market dynamics will be shaped. The generated market dynamics can then be validated by a variant of the event study method.

The major advantage of this design is its flexibility in both scenario design (e.g., deciding properties of the target commodity and the flow of events) and market dynamics generation (in our system it could be easily changed if new market agents are introduced). This generic commodity trading simulation platform will serve as the center piece for our future research in this area. We are especially interested in the following topics:

• Multi-agent framework for commodity trading simulation.
  In which we would formalize the construction of a commodity trading simulation with high fidelity.
• Can trading be learned ? -- Evidence from a commodity trading simulation.
  In which we would investigate whether it is possible to distinguish the intrinsic and extrinsic properties of a human trader. The measurement of these properties will be conducted on our trading simulation platform.
• Human-agent interaction in an event-based trading environment.
  In which we would investigate the effect of having both human and agent traders in an event-driven environment.

Market-Based Resource Allocation

In this line of research, we are interested in installing market mechanisms to guide resource allocation in a decentralized way. Depending on the properties of individual agents, resource constraints, and how uncertain we are about the problem properties, the market effectiveness might differ greatly. The research is thus focused on how we could design effective agent strategies or market mechanisms in various different configurations.

The first topic I am currently working on in this area is the design of robust market mechanisms that are capable of producing ideal allocation plans (of course, the definition of "ideal" allocations would differ from case to case) even in face of uncertainties like service disruptions or dynamic task arrivals. A number of approaches are currently being investigated, including the tweaking of mechanism parameters, aggregating multiple planning periods, and introducing after-market for emergent exchange. In some cases, if uncertainties are originated from the behaviors of other participating agents in the market, we might be able to better cope with such uncertainties via incremental observations.

The second topic I am interested in is on the agent-side, i.e., the design of agent strategies when the market rules are fixed. A number of stylized scenarios have been proposed in the research community, most notably the annual Trading Agent Competition, and this provides an ideal platform for a more comprehensive study on the design and the analysis of agent strategies.

• Robust mechanism design.
• Trading agent design and analysis.

Taxi is an important class of public transportation that provides convenience and flexibility levels close to that of car ownership. Unfortunately, taxi services are also very inefficient (50% idling time is quite common for a typical taxi fleet) since significant fraction of taxi services is delivered by roaming on the street. This inefficiency is most evident when large amount of demands suddenly emerge in a small area. Taxi service providers are usually slow in adjusting to this type of surge, and this results in unnecessary waiting on both driver-side and rider-side and greatly hurts operational efficiency.

In this project, we would address the issue of taxi service and demand mismatches in an urban environment. We will first develop the behavioral model for taxi drivers near the point of demand surge, and based on this model, we plan to propose practical mechanisms the could help to improve the quality of service (QoS) for the studied taxi fleet. All our studies will be based on the real-world data.

• Develop technique for processing and analyzing noisy real-world data.
• Modeling driver's queueing behaviors.
• Propose and test mechanism that would improve taxi fleet efficiency.