FinVerse: A New Benchmark for Financial Time-Series Forecasting

A new benchmark called FinVerse has been launched by researchers to enhance time-series forecasting in the finance sector, allowing for a more authentic evaluation of models. This benchmark overcomes the shortcomings of current benchmarks that rely on uniform error-based metrics across diverse series, which may not accurately represent real-world decision-making scenarios. For example, in stock forecasting, determining the direction of price movement is often more critical than simply reducing point-wise error. The FinVerse dataset comprises 116,897 financial time series with a total of 171.1 million observations, including 60,232 series with 17.4 million observations in its initial release. This benchmark is intended to facilitate a more realistic assessment of time-series foundation models in financial applications.