Investors often use their views of future GDP as an input when forecasting stock and bond market returns. It seems reasonable to turn this relationship around and argue that the bond, stock and real estate markets, which quickly incorporate investors’ current expectations about the future, might contain useful signals regarding the economy’s future direction. In fact, the Conference Board’s Leading Economic Index contains both a stock market and a bond market signal as leading indicators.
Using market data since 1970, PGIM’s Institutional Advisory & Solutions (IAS) examines several market signals and their ability to explain and predict next year’s GDP and the change in GDP from the previous year. The slope of the Treasury yield curve, the change in the Aaa-Baa quality corporate spread, the change in stock market valuation (CAPE) and real estate cap rate spread, carry information about future GDP growth, both in level and change. Fitting the market signals to the full data period, the signals explained 48% of the variation in next year’s GDP and 44% of the variation in the change in GDP from previous year.
However, what matters more to investors is the predictive power of market signals. Do bond, stock and real estate markets provide useful signals for forecasting future GDP? In terms of prediction, where only contemporaneous market signals are used to form forecasts, the predictive power of market signals has been poor. Relative to other market signals, the Treasury yield curve slope, stock market returns and the change in S&P 500 CAPE have exhibited better predictive power. However, even their absolute predictive power has been relatively low, and this power has fluctuated over time. To improve the predictive power of market signals, we explore combining many signals, and selecting them dynamically. Nevertheless, the average prediction error (RMSE) for next year’s annual GDP was 2.05% (in annual GDP percentage points), and 2.45% for the GDP change. Relative to an average GDP value of 2.4% and an average GDP change of 0%, this is a relatively noisy predictive signal.
Average In-Sample R2 and Prediction RMSE for Full Regression and Stepwise Regression