We're using cookies, but you can turn them off in your browser settings. Otherwise, you are agreeing to our use of cookies. Learn more in our Privacy Policy

building-capital-markets
THEME: CAPITAL MARKETS
2 July 2025 Research Foundation

Macroeconomic Drivers of Stocks and Bonds

Friedrich Baumann , Abdolreza Nazemi , and Frank J. Fabozzi, CFA

This CFA Institute Research Foundation brief discusses how machine learning can identify macroeconomic drivers of the shifting stock–bond correlation, offering actionable insights for asset allocation and risk management.

Macroeconomic Drivers of Stocks and Bonds View PDF
Macroeconomic Drivers of Stocks and Bonds Cover

Overview

Traditional investment wisdom holds that stocks and bonds tend to move in opposite directions, thus offering investors a natural hedge within balanced portfolios. But this longstanding negative correlation has shown signs of reversal, particularly amid recent inflationary pressures and monetary tightening. This shift raises a fundamental question: How should investors adapt their strategies when the traditional diversification benefits of stock–bond portfolios weaken or disappear?

“Macroeconomic Drivers of Stocks and Bonds ” answers that question by offering a data-driven framework to understand and forecast stock–bond return correlations under evolving macroeconomic conditions. Instead of relying on economic intuition or theory-based selection of drivers, this CFA Institute Research Foundation brief provides a practical, evidence-based framework for understanding stock–bond correlations in an era of macroeconomic transition.

By combining large-scale macroeconomic data with modern machine learning (ML) techniques, the authors offer a replicable method for forecasting correlations and identifying their key drivers. The findings have timely implications for asset allocation, risk management, and investment strategy.

Research Foundation Ad
Support Professional Excellence

Support Research for the Global Investment Community. Your donation raises professional excellence and enables the foundation to continue its important mission to fund, publish, and distribute in-depth, high quality, independent research.

The Breaking of the Stock–Bond Link

Historically, monetary policy has served as a powerful stabilizing force, anchoring the inverse relationship between equity and bond returns. But since the COVID-19 recovery, inflation has surged, the Federal Reserve has tightened aggressively, and macroeconomic uncertainty has escalated. These conditions have pushed the stock–bond correlation back into positive territory — something not seen consistently for more than two decades. In this environment, the authors posit, traditional portfolio construction frameworks face significant challenges.

A New Approach

Instead of picking variables based only on economic theory, the authors use a method called stability selection. This approach runs a special type of statistical selection many times on slightly altered data to find which macroeconomic indicators consistently matter most. The authors’ dataset spans the 1959–2023 period and includes 127 monthly macroeconomic indicators from the Federal Reserve Economic Data (FRED) database.

After identifying the most reliable predictors, the researchers use an ML method called random forests to predict future stock–bond correlations and figure out which variables have the biggest impact. They demonstrate the results are not only statistically strong but also economically intuitive: indicators tied to manufacturing activity — such as nondurable goods employment and new manufacturing orders — emerge as critical predictors.

Key Findings and Insights

  1. Stock–Bond Correlation Is Not Constant: The brief confirms that the stock–bond correlation has shifted dramatically over time, from positive (1930s–1950s), to negative (2000s–2020), and back toward positive in the post-COVID era.
  2. Leading Manufacturing Indicators Are Predictive: Across multiple modeling — including linear regression, principal component analysis (PCA), and random forests — variables related to manufacturing consistently show high importance.
  3. Machine Learning Outperforms Traditional Models: Out-of-sample forecasting using random forests significantly outperforms linear regression, reducing root mean square error. This reinforces the value of ML for capturing nonlinear relationships and interactions in complex macroeconomic data.
  4. Dimensionality Reduction Aids Interpretation: Grouping variables into thematic categories (e.g., manufacturing, housing, financial markets) via PCA improves interpretability and highlights economic linkages, especially when applied in tandem with stability selection.
  5. Some Long-Used Indicators Underperform: Surprisingly, traditional variables such as the S&P 500 Index and inflation rates were less predictive of future stock–bond correlations compared with manufacturing-related indicators.

Practical Applications

Applying insights from the paper, investors can:

  • Reassess diversification assumptions. In periods when stock and bond returns move together, investors must rethink the risk-hedging role of bonds in their portfolios. Strategies that assume a persistent negative correlation may fail during inflationary or tightening cycles.
  • Integrate leading economic indicators. Monitoring key manufacturing and housing signals can provide early warnings of changes in the stock–bond dynamic. These indicators offer a forward-looking lens for portfolio construction.
  • Use ML judiciously. Tools like random forests can uncover nonlinear relationships missed by traditional models. While they require more data and computing power, they also offer improved predictive accuracy and robustness.
  • Adapt risk models and stress tests. Risk models should incorporate the possibility of correlation regime shifts. Stress testing under varying stock–bond correlations can help identify vulnerabilities in multi-asset portfolios.

    A Toolkit for Modern Portfolio Strategy

    Understanding the changing relationship between stocks and bonds is a growing challenge, especially now that their returns are moving together more often than not.

    This brief helps investors move beyond intuition by offering a robust, ML-powered framework that can adapt to complex economic regimes — ensuring more-informed decision making in a rapidly evolving financial landscape. The brief gives investment professionals a practical, data-driven framework to tackle that problem.

    By grouping related economic indicators into themes and ranking their importance, the researchers offer a clear, accessible approach to analyzing complex macroeconomic data. The result: a reliable toolkit for investors to better predict stock–bond dynamics, adapt portfolio strategies, and manage risk in a shifting economic landscape.