Description

While simple univariate index forecasting often yields disappointing results in live trading, cross-sectional approaches unlock deeper alpha potential. We introduce an applied framework for financial time series forecasting using Chronos-2, a state-of-the-art pretrained probabilistic model. Following the framework of Fischer and Krauss (2018), we build a cross-sectional daily market-neutral Long/Short portfolio of individual equity assets.

Using a 30-year historical dataset of individual S&P 500 constituents sourced via yfinance and structured to eliminate survivorship bias, this course demonstrates how to transform historical raw returns into predictive distributions. We evaluate the performance of a daily rebalanced portfolio against the S&P 500 benchmark (accounting for commissions and execution slippage). Lastly, we theorize further methods to enhance the strategy by integrating Empirical Mode Decomposition (EMD) to filter market noise and isolate high-value decomposed technical covariates (Guo et al., 2024).