The Taiwan Banker

Prediction markets have experienced a renewed surge of attention following a series of legal and geopolitical developments. A 2024 US federal court decision allowed platforms such to expand their event contracts beyond narrowly defined economic indicators into broader real-world outcomes, including political and policy-related questions. This ruling was widely interpreted as a turning point, signaling that certain prediction market products could be treated as legitimate financial instruments rather than prohibited forms of gambling. More recently, rising geopolitical tensions have driven a sharp increase in trading volumes on geopolitical event contracts. These markets, which attempt to price the probability of events such as military escalation or sanctions, have attracted both institutional curiosity and retail participation. This upheaval has also exposed a fundamental vulnerability, however: the potential for insider trading. Reports of large, preternaturally accurate trades placed shortly before key geopolitical developments – often by new accounts – have raised concerns that individuals in the government, intelligence, or defense sectors with privileged information may be using prediction markets to monetize non-public knowledge. Insider trading is not an abuse of this structure; it was always understood to be a feature, or at least not a bug. In a 2007 paper entitled Insider Trading and Prediction Markets, Robin Hanson, an economics professor at George Mason University, wrote that “Many have argued that insider trading laws encourage investment in public corporations by assuring investors that they are not trading against other investors with vastly superior information. I will suggest that this benefit, even if real, now comes at the cost of discouraging innovation in our corporate informational institutions, and that this is a needless cost, since there are better ways to encourage both investment and institutional innovation.” Hanson is considered the architect of prediction markets in the current form. While this passage addresses publicly traded corporations rather than the government, as a general principle in economics, any cost is also an incentive. The current fear is that insider trading could be contributing to policy volatility with the specific purpose of creating trading opportunities. In response to recent incidents, US legislators have introduced proposals to address the threat of insider trading, in part by banning entire domain areas, like geopolitics; but multidimensional competition between structural models is precisely where schemes to quantify uncertainty could add the most value. Consider the use cases for model quantifying the risk of a conflict involving China, to the best ability of the top forecasters. This is the concept of superforecasting, popularized by Wharton professor of political psychology Philip Tetlock. Such data could be integrated into financial models to guide budgetary decisions, but its real value comes from its integration with conditional markets. Markets on questions of the form “likelihood of outcome B given precondition A” facilitate the formation of hypotheses on the cause of outcome B, in order to prevent (or encourage) its occurrence, rather than just reacting. Superforecasting de-emphasizes fast-moving insider knowledge in favor of careful logical deduction. It was originally envisioned to enable discourse, rather than market trading. In contrast to Hanson’s vision of pulling together disparate pieces of partial knowledge, here, participants are encouraged to explain their reasoning, which would tend to discourage claims without legitimate sourcing. Tetlock argued that the benefits of domain-specific expertise may not outweigh a big-picture understanding. Nevertheless, the most fundamental implementation difficulty with superforecasting platforms has been a lack of liquidity – particularly for the most interesting conditional markets. To create social value, they should have a broad bank of questions, particularly considering that valid markets must have clear resolution criteria, and it may not always be clear ax ante which questions need to be asked. A “black swan” event, by definition, was never even modeled in the first place; on a semantic level at least, modeling more questions prevents the occurrence of black swans. Furthermore, interest in answering questions is itself a source of information, and the signal of market exit is lost without real-world incentives. Superforecasting platforms like Metaculus operate on a demonetized model, which eliminates the possibility of insider trading, but which also forgoes these benefits. Complete illiquidity appears to be sub-optimal, but limits on total market or position size also appear to be necessary, depending on the sensitivity of the topic. Here it is worth comparing prediction markets to the sports betting platforms, which have become mainstream since the 2018 U.S. Supreme Court decision in Murphy v. NCAA, which struck down the Professional and Amateur Sports Protection Act (PASPA). This ruling allowed individual states to legalize and regulate sports betting, leading to rapid market expansion, widespread commercialization, and sophisticated risk management practices by operators. One interesting aspect of gambling platforms is their active use of position size limits for management. Unfortunately, the purpose is frequently to give the house an unfair advantage, banning “whales” who often make large, accurate wagers. An ideal market, on the other hand, could link position limits positively with reputation based on the outcomes of previous positions – creating a partially monetized system which discourages both impulsivity and over-reliance on one-time inside views. In principle, it might be possible to segment gaming from other markets for separate regulation based on the event category. This is the current approach, although loopholes reduce its effectiveness. The larger point, however, is that these constraints – market liquidity plus pro-social regulation by position limits – implies a legal category for markets with research functionality, which does not yet exist directly. Prediction markets look more like equity than traditional insurance contracts; yet they have a stronger price discovery function than stock markets, which exist primarily to channel financial capital into real-economy investments. Their infrastructure will require regulatory innovation just as much as technical. In this way, the scale of their disruption can also be compared to the emergence of digital assets over the past decade – which likewise blurred the boundaries between securities and currencies, while enabling entirely unexpected forms of cyber threats. Despite that similarity in overall scale of disruption, while cryptocurrencies have pushed the talent requirements of the financial industry away from its roots in accounting more toward IT, prediction markets will herald a return toward a deeper understanding of economics. Their effectiveness will only be understood through a combination of theory with statistical analysis, and their design objectives will only be understood at a certain level of abstraction. Importantly, such objectives may not all lead It might also be useful to note the potential for progress of a technological nature, particularly if a legislative solution is intended to remain valid for a decade or more. The most important role of AI will likely be liquidity provision. It struggles with out-of-sample inference, and will probably prefer a more technical trading style for some time to come, but the resulting short-term churn can create opportunities for human traders who can truly think outside the distribution.