The application of prediction markets to project prioritization in the not-for-profit sector

Grainger, Daniel Ali (2017) The application of prediction markets to project prioritization in the not-for-profit sector. PhD thesis, James Cook University.

[img]
Preview
PDF (Thesis)
Download (2MB) | Preview
View at Publisher Website: https://doi.org/10.4225/28/5ac41aeb4ac38
 
38


Abstract

The principal objectives and scope of the study: The goal of this thesis is to ultimately propose how prediction markets may be used to select the best possible portfolio of projects in a not-for-profit setting. The first logical principal objective is to identify a quality signal that demarcates a good from a bad market prediction (or decision). Upon doing so, the second principal objective of configuring prediction markets (into a decision market) to decide if a project should be in or out of the project portfolio, and importantly the probability that this selection is the best possible one, is achievable. This objective is achieved via a novel theoretical model introduced in this thesis and by three investigations to test the theory. The third and final principal objective is to consider synthesize the finding of this thesis to propose how to improve current prediction markets and create a new type of prediction and decision market that is augmented or embedded with the quality signal identified by the thesis theory and investigations.

Due to time and budget constraints the scope of this thesis does not extend to building a real-world implementation of the prediction and decision market types introduced by this thesis for real-world organizations. However, at the time of writing a joint venture arrangement between the university and other stakeholders is in development to implement this thesis' decision market for project selection in possibly two not-for-profit organizations.

The methodology employed: The keystone and novel contribution of this thesis is the identification of a quality signal that can be used to improve prediction and decision markets. The methodology to achieve this is to first establish a theoretical model that builds upon the work of previous research but that provides an original contribution in its focus on quality signals in prediction markets.

The idealized theoretical model suggests a possible quality signal candidate denoted as relevant information level i.e. the proportion of the market bids that are conditioned on private information. Given relevant information is arrived at theoretically, it is important to robustly test the hypotheses that increasing relevant information increases the probability of prediction and decision markets attaining the best possible (fully informed) predictions and decision respectively.

The first investigation is a computer simulated control-treatment setup in which each control and treatment prediction (and decision) market is identical in every way except for the relevant information level. The statistical significance of relevant information level is then analyzed. The benefit of computer simulations is the large number of prediction and decision market games that can be run. The weakness is that all traders are rational. Therefore, the second investigation incorporates, into the prediction and decision markets, human participants.

The second investigation involves a prediction market web-game created by the PhD candidate. Multiple prediction market web-games are run in a control-treatment setup in which relevant information is allowed to vary with all else being held the same for each control and its treatment. The statistical significance of relevant information level is then analyzed. The strength of this investigation is the incorporation of human idiosyncrasies; however, the weakness is this investigation remains within the confines of the laboratory setting. Hence the third and final investigation involves the analysis of real-world prediction market data.

The analysis of real-world prediction market data is undertaken to extend the testing of the hypotheses into a real-world setting. The strength of this is that the hygienic conditions of the laboratory are removed, however, the weakness is the potential endogeneity and confounding problems that can arise. To counter these potential problems, a control function approach is used to control for endogeneity and a fine strata continuous propensity score approach is used to control for confounding.

Summary of the results: The new theoretical model finds relevant information level to be a sufficient and necessary condition for prediction and decision markets to attain the best possible predictions and decision respectively. The computer simulations find that increasing relevant information level is a statistically significant effect that leads to improving the probability of prediction and decision markets attaining the best possible predictions and decisions. The prediction market web-games with human participants also confirms that relevant information is a statistically significant effect that when increased leads to increasing the probability of attaining the best possible prediction. Finally, the analysis of real-world prediction market data confirms that relevant information is a statistically significant effect that when increased leads to increasing the probability of attaining best possible decisions.

The principal conclusions: There are several conclusions able to be drawn from this thesis as follows:

Relevant information level is robustly tested in this thesis as a quality signal for prediction and decision markets to demarcate good prediction and decisions from bad ones. This is the key contribution by this thesis to research on prediction and decision markets. Prior to this work there existed no metric to assess the quality of a prediction or decision emanating from these markets. Without such a 'quality' metric, confidence in the associated predictions and decisions would remain logically unjustified. Hence 'relevant information level', as a quality signal, proposed and robustly tested by this thesis brings with it the ability to justly demarcate a good prediction or decision from a bad one.

Current real-world prediction markets may be simply augmented with a publicized measure of their relevant information level to improve markets in much the same way that Akerlof quality signals do so. That is, the implication of this thesis does not require radical changes to current real world prediction and decision markets. Rather, by simply publicising the 'relevant information level' metric, the confidence warranted in a prediction or decision market is revealed. In much the same way that Akerlof quality signals were embodied in small changes (e.g., second-hand car guarantees), 'relevant information level' as a quality signal is a small change i.e. publicising and therefore a guarantee of the efficacy of a prediction or decision market.

The new prediction and decision markets that were built in this thesis have the potential to select the best possible portfolio of projects. With high quality decision market selection comes the confidence that the best possible selection is made. The selection of the best possible portfolio of projects is a logical extension to this feature. Establishing a means of measuring how well prediction and decision markets select the best possible portfolio of projects is an important and novel contribution by this thesis. The ultimate goal of confidently guiding project selection so as to best leverage scare economic resources is made possible in this thesis.

Item ID: 53034
Item Type: Thesis (PhD)
Keywords: information market, decision market, prediction market, joint elicitation, long term projects
Related URLs:
Additional Information:

Publications arising from this thesis are available from the Related URLs field. The publications are:

Chapter 3: Grainger, Daniel, Sun, Sizhong, Watkin-Lui, Felecia, and Case, Peter (2015) A simple decision market model. Journal of Prediction Markets, 9 (3). pp. 41-63.

Date Deposited: 04 Apr 2018 04:24
FoR Codes: 14 ECONOMICS > 1401 Economic Theory > 140103 Mathematical Economics @ 50%
14 ECONOMICS > 1403 Econometrics > 140303 Economic Models and Forecasting @ 50%
SEO Codes: 91 ECONOMIC FRAMEWORK > 9104 Management and Productivity > 910406 Technological and Organisational Innovation @ 50%
91 ECONOMIC FRAMEWORK > 9102 Microeconomics > 910206 Market-Based Mechanisms @ 50%
Downloads: Total: 38
Last 12 Months: 4
More Statistics

Actions (Repository Staff Only)

Item Control Page Item Control Page