Abstract not found

Future market conditions can be a pivotal factor in making business decisions. We present and evaluate methods used by our agent, Deep Maize, to forecast market prices in the Trading Agent Competition Supply Chain Management Game. As a guiding principle we seek to exploit as many sources of available information as possible to inform predictions. Since information comes in several different forms, we integrate well-known methods in a novel way to make predictions. The core of our predictor is a nearest-neighbors machine learning algorithm that identifies historical instances with similar economic indicators. We augment this with an online learning procedure that transforms the predictions by optimizing a scoring rule. This allows us to select more relevant historical contexts using additional information available during individual games. We also explore the advantages of two different representations for predicting price distributions. One uses absolute prices, and the other uses statistics of price time series to exploit local stability. We evaluate these methods using both data from the 2005 tournament final round and additional simulations. We compare several variations of our predictor to one another and a baseline predictor similar to those used by many other tournament agents. We show substantial improvements over the baseline predictor, and demonstrate that each element of our predictor contributes to improved performance.

Abstract. The TAC SCM tournament is moving into its fourth year. In an effort to track agent progress, we present a benchmark market efficiency comparison for the tournament, in addition to prior measures of agent competency through customer bidding. Using these benchmarks we find statistically significant increases in intratournament market efficiency, whereas agents are generally decreasing in manufacturer market power. We find that agent market share and bid efficiency have increased while the variance of average sales prices has been significantly reduced. Additionally, we test for a statistical relationship between agent profits and the bullwhip effect. 1

Abstract. Supply chains are ubiquitous in the manufacturing of many complex products. Traditionally, supply chains have been created through the interactions of human representatives of the companies involved, but advances in autonomous agent technologies have sparked an interest in automating the process. The Trading Agent Competition Supply Chain Management (TAC SCM) scenario provides a unique testbed for studying supply chain management agents. This paper introduces TacTex-05, the champion agent from the 2005 competition, focusing on its ability to adapt to opponent behavior over a series of games. The impact of this adaptivity is examined through both analysis of competition results and controlled experiments. 1

The Supply Chain Trading Agent Competition (TAC SCM) was designed to explore approaches to dynamic supply chain trading. During the course of each year’s competition historical data is logged describing more than 800 games played by different agents from around the world. In this paper, we present analysis that is focused on determining which features of agent behavior, such as average lead time or selling price, tend to differentiate agents that win from those that don’t. We begin with a visual inspection of games from one bracket of the 2006 semi-final rounds. Plots from these games allow us to isolate behavioral features which do, in fact, distinguish top performing agents in this bracket. We introduce an information gain based metric that we use to provide a more complete analysis of all the games from the 2006 quarter-final, semi-final and final rounds. The technique involves calculating the amount of information gained about an agent’s performance by knowing its value for each of 20 different features. Our analysis helps identify features that differentiated winning agents. In particular we find that, in the final rounds of the 2006 competition, winning agents distinguished themselves by their procurement decisions, rather than their customer bidding decisions. We also discuss how the information gain analysis could be extended by agent developers to identify potential weaknesses in their entry.

We present basic building blocks of an agent that can use observable market conditions to characterize the microeconomic conditions of the market and predict future market trends. The agent can use this information to make both tactical decisions such as pricing and strategic decisions such as product mix and production planning. We develop methods that can learn dominant market conditions, such as oversupply or scarcity, from historical data using computational methods to construct price density functions. We discuss how this knowledge can be used, together with real-time observable information, to identify the current dominant market condition and to forecast market changes over a planning horizon. We validate our methods by presenting experimental results in a case study, the Trading Agent Competition for Supply Chain Management. 1

Abstract. This paper proposes a novel method to characterize the performance of autonomous agents in the Trading Agent Competition for Supply Chain Management (TAC-SCM). We create benchmarking tools that manipulate market environments to control the conditions under which we test trading agents. Using these tools, we show how developers can inspect their agents and unveil behaviors that might otherwise have gone undiscovered. 1

was designed to explore approaches to dynamic supply chain trading. During the course of each year’s competition historical data is logged describing more than 800 games played by different agents from around the world. In this paper, we present analysis that is focused on determining which features of agent behavior, such as the average lead time requested for supplies or the average selling price offered on finished products, tend to differentiate agents that win from those that do not. We present a visual inspection of data from 16 games played in one bracket of the 2006 TAC SCM semi-final rounds. Plots of data from these games help isolate behavioral features that distinguish top performing agents in this bracket. We then introduce a metric based on information gain to provide a more complete analysis of the 80 games played in the 2006 TAC SCM quarter-final, semi-final and final rounds. The metric captures the amount of information that is gained about an agent’s performance by knowing its value for each of 20 different behavioral features. Using this metric we find that, in the final rounds of the 2006 competition, winning agents distinguished themselves by their procurement decisions, rather than their customer bidding decisions. We also discuss how we used the analysis presented in this paper to improve our entry for the 2007 competition, which was one of the six finalists that year.

In many multiagent settings, each agent’s goal is to come out ahead of the other agents on some metric, such as the currency obtained by the agent. In such settings, it is not appropriate for an agent to try to maximize its expected score on the metric; rather, the agent should maximize its expected probability of winning. In principle, given this objective, the game can be solved using game-theoretic techniques. However, most games of interest are far too large and complex to solve exactly. To get some intuition as to what an optimal strategy in such games should look like, we introduce a simplified game that captures some of their key aspects, and solve it (and several variants) exactly. Specifically, the basic game that we study is the following: each agent i chooses a lottery over nonnegative numbers whose expectation is equal to its budget bi. The agent with the highest realized outcome wins (and agents only care about winning). We show that there is a unique symmetric equilibrium when budgets are equal. We proceed to study and solve extensions, including settings where agents must obtain a minimum outcome to win; where agents choose their budgets (at a cost); and where budgets are private information.

This paper proposes a novel method to characterize the performance of autonomous agents in the Trading Agent Competition for Supply Chain Management (TAC-SCM). We create benchmarking tools that manipulate market environments to control the conditions and provide guidelines to test trading agents. Using these tools, we show how developers can inspect their agents and unveil behaviors that might otherwise have gone undiscovered.