Friday, October 6, 2017

Manipulation in Prediction Markets - Chasing the Fraudsters

Summary by Keith Robinson Jr.


In this article, the researchers address the issues of manipulation and fraud in prediction markets and examine fraud detection approaches. While the authors acknowledge the versatility and forecasting accuracy of prediction markets in comparison to polls or even statistical models, they are not without issues. First, the researchers concluded that to some extent, prediction markets can be manipulated (manipulation defined as a speculative attack that achieves its objective of changing prices). Researchers have come to mixed conclusions regarding manipulation in prediction markets. While anecdotal evidence has shown that the manipulation affects the information aggregation aspect, not reducing predictive accuracy of forecast, other evidence has revealed that "manipulators highly incited for inaccurate predictions, can diminish the predictive power of the markets down to a level that is no better than random guessing" (p. 2981).

Second, the literature looks at fraud in prediction markets. Frequently, extant literature on prediction markets consider traders playing by the rules; however, traders may play by the rules and create other ideas how to cheat the market. They focus on the visualization of the Fraud Cube (see Figure 1.), a framework to understand and uncover where a prediction market may be manipulated or cheated. In order fraud to occur in prediction markets three dimensions must occur: 1) desire/objective (whether to disrupt the market, self-enrichment, or both, 2) temporal horizon (short-term or in the long run - realize quick profits or destroy market prediction or vested interest in the outcome or decisions derived), and 3) source of incentive (the incentive is caused by an inner incentive scheme inside the market or externally).

Figure 1. Framework to understand and uncover where a prediction market may be manipulated or cheated. 

Next, existing fraud detection and trading patterns recognized by Blume et al. (2010) are highlighted. Prominent detection strategies revolve around "ping-pong indicators," focusing on transfer or money, and prominent-edge indicators," taking a look at stocks. Realizing Blume's indicator shortcomings, the researchers developed a simple algorithm, easily applicable for practitioners. This algorithm utilizes a scoring system, scoring traders with "suspicious points" whereas the top ranks have the highest probability to commit prediction market fraud. The researchers evaluate the algorithm during a 12 week data collection encompassing 2,111 participants conducting 112,386 transactions. The algorithm was able to find 484 suspects, 6 traders having more than 200 suspicious points with 551 points being the highest yield.


The authors utilized existing literature to simplify a method for application by practitioners. While it is not an end all be all solution, the authors acknowledge their algorithm's own limitations. All in all, the article highlights an under-addressed issue that can damper prediction market accuracy and implications for such.

Source: Kloker, Simon & T. Kranz, Tobias. (2017). Manipulation in Prediction Markets - Chasing the Fraudsters. Retrieved from


  1. I agree with you Keith that this algorithm is not an end all be all solution. Though the authors identify certain attack patterns that fraudsters might use, there are other creative attacks that are still not detected. Though the aim of the study was to create a heuristic approach that is easily applicable to practitioners, this study has many limitations as the author’s addresses. Despite this the authors did introduced an algorithm to spot fraud in prediction markets, which was capable of creating a valid list of suspects that does not rely on complicated methods.

  2. I also agree with you Keith and Michael about how the algorithm is not an encompassing solution. While producing an algorithm to produce a list of valid suspects is helpful at the overall effectiveness of using prediction markets. I would like to raise the question of is there some steps prior to improve trader selection, so that there would be less of a risk of fraud in the first place. Or would eliminating certain candidates from the trading pool, would that ruin the validity of the prediction market or just force fraud in a way not detectable by an algorithm.

  3. This is quiet interesting as I was not aware that the market price in a prediction market could be subject to manipulation. Ultimately, if this ends being true then prediction markets have no future. Nonetheless, I feel as though you did not clearly explain how the market price is manipulated. Could you clarify?

  4. It seems the algorithm was able to do a decent job of producing a list of suspects for prediction markets. I agree that there are limitations to it and that it cannot answer it all. Overall it seems the algorithm did spot fraud in prediction markets reasonably but more research must likely continue with it.

  5. I'm curious to see if this algorithm could be replicated in a real world scenario. If the researchers want this algorithm to be used in actual prediction markets, replicated studies would give it more validity. Do you know how they set up their equation with the point system they developed?