The Role of a Messaging System in Initial Online Freelance Market

Abstract

Online freelance market is made up of two major players: Buyers and Sellers. After joining the website of an online freelance company, a buyer would post a project with the detailed description about the work or the explanation about an ideal seller. After a buyer finishes posting her projects, potential sellers take part in the competition to win the business deal. The whole procedure of a transaction takes a reverse auction process. In other words, sellers write out a bid which includes bid price and bid period. (bid price refers to how much sellers would charge for carrying out the project and bid period refers to the expected time required to complete the project.) After examining all the bid-specific information, the buyer would choose the winner of the auction. In this process, market players, especially buyers, may encounter serious information asymmetry problem. The seller who suggests a bid price has more information about the final output of the project than a buyer has. That being said, sellers who have more inside information also have a chance to give a bid in their favor. The inside information would be the true capability to carry out the project and/or the willingness to put effort into the project. An important issue in this market would be the possible imbalance of the information. In economic context, the lemon problem from information asymmetry may drive out market players who would be so diligent, sincere, and competent that they could contribute market-settlement. The informational asymmetry may result in a market that retains only the players having low level of capability, accelerating more exit of market players. This vicious circle is a matter of life and death to the operator of the online freelance service. So it is very important for the online freelance operator to relieve the information asymmetry problem. Online freelance companies can consider introducing a messaging system within the online auction process. Through the messaging system, sellers (freelancers) have an opportunity to appeal themselves about their willingness to do a project and to provide the information about the ability they have for conducting a project. Therefore, adopting the messaging system seems to be an appropriate solution to relieve the information asymmetry problem that online freelance companies have. But the actual effect of message in the decision process of buyers (clients) has not been explored much. This study builds a structural dynamic model of utility maximizing behaviors of buyers and estimates the model using a large dataset obtained from a Korean online freelance website. Buyers are modeled to be dynamic utility maximizer in the sense that a buyer, in each time period, has to make a decision on whether to select a bid or to wait for another bid. If more promising bid is expected to come, the buyer may wait for future bid. Otherwise, the buyer may select a bid among the currently available ones. In order to incorporate such feature in buyer behaviors, this study utilizes the finite horizon dynamic programming approach in modeling explicitly the buyer behaviors. The data set comes from a domestic online freelance marketplace. An interesting feature of the site is that sellers can send messages to buyers and the message data are stored in the website database. This study explores the role of the message system of the site in reducing the information asymmetry problem..In terms of estimation, we first estimated a model that does not incorporate the message into the utility function. Inthe model without the message data, buyers turned out insensitive to bid price. Note that if one does not control for the product quality in the demand function where the quality indeed affect the demand and at the same time is correlated with price, the price effect could be underestimated in the model without the quality. This typical omitted variable bias may kick in the online freelance market problem if the bid price indeed reflects the true capability of the sellers. That is, the sellers with a higher level of capability may post higher prices and the buyers may take such capabilities into consideration when marking decision. If that is the case, the quality of bid is observed by the decision making buyers but not by researchers. To control for the unobserved quality, we include the message data into out model. If the messages from the sellers contain the information that buyers can utilize to assess the capability of the bids, the model with the message data can better identify the true effect of bid price. In order to incorporate of the contents of the message into the model, we transformed the text information in the message dataset into numeric data via a topic modeling approach. In the model with the message data, the buyers are estimated to be more price sensitive than in the model without the message information.