The Journal of Prediction Markets

Predicting the Outcome of Indian Super League Football Matches Using Machine Learning

Sweta Tripathy — 2026-06-03

The study is an extensive examination that aims to forecast the critical factors impacting football match outcomes, with a particular focus on win/loss determination in the Indian Super League (ISL). The study makes use of three different machine learning algorithms: AdaBoost, Support Vector Machine (SVM), and Classification and Regression Trees (CART), using data from a total of 377 matches. The main goal of the study is to identify and select the most influential features that influence a match's outcome. Through the use of these algorithms, we want to increase forecast accuracy and offer insights into the crucial elements that may be strategically controlled to boost team performance.

The study assesses each algorithm's prediction power and analyzes how well it performs in feature selection. The findings show notable differences in feature importance between the models, highlighting the advantages and disadvantages of each model. This work contributes to the wider use of machine learning in sports analytics while also deepening our understanding of football performance drivers.

General Regression Neural Network Based Forecasting for Indian Exports of Beauty and Skin Care Products

Manas Tripathi — 2026-06-03

The export of beauty and skincare products from India to international markets is rapidly growing with the rise in the global demand for natural, herbal, and Ayurvedic solutions. However, the beauty and skincare sector is volatile and unpredictable due to changing consumer preferences, beauty trends, global competition, and trade policies. Accurate forecasting of Indian beauty and skincare exports is essential for practitioners to optimize business strategies and maintain India’s competitive edge in this growing market. The present study compares the prediction performance of GRNN, MLP-ANN, and ARIMA methods in forecasting export volumes. The study utilizes historical export data from 2007 to 2024 to analyze trends and patterns in India’s beauty and skincare exports. The findings demonstrate that the machine learning model, GRNN, outperforms MLP-ANN and ARIMA, in capturing complex, non-linear data, resulting in more accurate and reliable export forecasts. This research provides valuable insights for policymakers, exporters, and businesses by offering precise predictions that can facilitate strategic decision-making, optimize supply chains, and support market expansion.

How accurate are forecasts on geopolitical events from human collectives?

Oliver Strijbis — 2026-06-03

Accurate forecasts of geopolitical events are essential for security, foreign, and macroeconomic policy. Among human-based forecasting methods, predictions of collectives have established themselves as particularly accurate and useful. In particular, prediction polls and prediction markets have become well-studied and established methodologies. This article evaluates the discrimination and calibration of a prediction market on geopolitical events conducted in 2023 and 2024. It makes two contributions to the literature. First, it is the first article to provide evidence of the forecasting accuracy of a real-money prediction market on geopolitical events. Second, it provides one of the first comparisons of a prediction market’s forecasting accuracy with those of prediction polls for geopolitical events. This way, it contributes to a still small but growing literature that tries to establish the conditions under which prediction polls or prediction markets generate more accurate forecasts.

Optimization strategies in portfolio management: Do they influence performance?

Debayan Chakraborty — 2026-06-03

The study assesses the relative out-of-sample performance of different portfolio optimization strategies across four mean-risk frameworks and a benchmark naïve (1/N) strategy using weekly price data of a stock index, foreign currency, gold, natural gas, and crude oil from December 1997 to December 2023. For each mean-risk framework, we employ two optimization strategies: risk minimization and Sharpe ratio maximization. Using various risk-adjusted and economic measures, the out-of-sample performance analysis of all the strategies suggests that the Sharpe ratio maximization strategy of the mean-CVaR framework is the best performing model, while the variance minimization model of the mean-variance framework performs worst.