This study investigates the accuracy of prediction market forecasts of economic variables by using new data from the prediction market Kalshi’s markets for oil prices and US Treasury yields. Specifically, three methods of calculating a forecast from the market data are proposed and compared to benchmark ARIMA models, and Long Short-Term Memory (LSTM) neural networks. The results show that the prediction market forecasts, especially when using Kernel Density Estimation Probability Smoothing as the calculation method, generally outperform ARIMA models and beat the LSTM models for the US Treasury 10Y yield market. Furthermore, if the prediction market forecasts are separated by trading volume, the ”High volume” category vastly outperforms ARIMA and LSTM models for both oil prices and US Treasury yields. The findings show that prediction markets as a forecasting tool for economic variables are competitive with even the most advanced time series forecasting algorithms, despite observed challenges of low market liquidity. This strengthens the argument for further adoption and legalization of prediction markets.