Using Wavelet Transformation to Predict S&P 500 Performance
Log in to collectAcademic paper
Stock Return Predictability in Frequency Domain
Jie Kang; Fuwei Jiang; Zhifeng Dai
- Central University of Finance and Economics
- Xiamen University
- ?Central University of Finance and Economics (CUFE)
- Changsha University of Science and Technology
Strategy in a nutshell
The strategy employs wavelet frequency domain decomposition (MODWT with Haar filter) to break stock returns into low-, mid-, and high-frequency components. Using the long-term (low-frequency) component, the authors construct an Aligned Macroeconomic Index (AMI) via Partial Least Squares (PLS) regression, aligning macroeconomic variables with predictive power for stock returns. The AMI is compared against principal component (PC) and equally-weighted (EW) factors, offering a superior, target-informed measure for forecasting.
Economic rationale
This approach integrates frequency-domain analysis and supervised dimensionality reduction to capture the true macroeconomic drivers of equity returns. Traditional predictors often contain noise or irrelevant components, limiting their predictive power. By aligning macroeconomic variables with long-term stock return behavior, the AMI isolates the latent economic factor influencing returns, enhancing both in-sample and out-of-sample forecast accuracy across market cycles.