EXAONE Forecast for Finance Model Introduced for Market Predictions

A technical report published on arXiv introduces EXAONE Forecast for Finance (EXAONE Finance), a specialized time series foundation model built for financial market forecasting. Unlike general-domain models that rely heavily on computationally expensive self-attention mechanisms, EXAONE Finance uses an attention-free architecture combining causal 1D convolution and a group-aware pooling multi-layer perceptron.

The model incorporates masked context augmentation during training to handle intermittent data points commonly found in financial markets. It was pretrained on a comprehensive financial corpus spanning equities, foreign exchange, commodities, crypto-assets, fixed income, and macroeconomic indicators.

According to the benchmark evaluations on FinVerse, EXAONE Finance achieved state-of-the-art results, ranking first across point-forecast accuracy, cross-sectional asset ranking, and portfolio profitability.