Analysis of Dividend Data Offers New View of Asset Pricing Puzzles
How are the real economy and financial markets related? The traditional consumption-based asset pricing model predicts that asset prices are determined by the covariance of future cash flows with consumption growth.
Using this model as a benchmark, the empirical macro-finance literature has documented three puzzling facts:
- stock returns are on average too high relative to bond returns (the equity premium puzzle)
- stock returns are too volatile relative to fundamentals (excess volatility puzzle), and
- stock returns are to some extent predictable.
To explain these puzzles, a new generation of asset pricing models has been developed relying on different economic mechanisms in terms of preferences or technology. However, all of the leading models predict that the asset pricing puzzles ought to be more pronounced for long-term assets.
Ralph S.J. Koijen, a CME Group Visiting Fellow in 2011–2012, and his colleagues provide a new perspective on these puzzles using prices of dividend strips. These are assets that pay dividends on the stock index for only a limited number of periods—say, for the next two years—unlike a stock index, which pays dividends in perpetuity.
In contrast to current leading models, their work with dividend strips found that
- expected returns on short-term assets (with a maturity of two years) are higher than on the index;
- short-term assets are more volatile than fundamentals, leading to excess volatility, yet
- they have strongly predictable returns compared to the index.
These asset pricing facts help identify the risks that investors care about. The key difference between short- and long-term assets is their exposure to temporary, or business-cycle, shocks. Short-term assets do poorly during severe economic downturns when firms temporarily cut back on dividends, while long-term assets are more exposed to shocks to trend growth. Business-cycle shocks hardly matter in mainstream asset pricing models, which has led to the conclusion that the welfare costs of business cycles are small—a line of work initiated by Robert Lucas. Koijen’s work suggests that the costs of business cycles may be non-trivial, at least in models with a maturity structure of asset prices they document.
In follow-up work, Koijen and colleagues broaden the evidence for these patterns, using a novel data set of dividend futures. This new data set has a richer maturity structure (up to 10 years) and has data that spans three major economic regions (the U.S., Europe, and Japan). The results for the U.S. were also seen in Europe and Japan.
The researchers then used these data to construct equity yields, the equivalent of bond yields for stocks. They show that equity yields can be decomposed into a risk premium and an expected growth component. Both components vary by maturity, which makes it possible to compute a term structure of growth expectations of dividends, GDP, and consumption by maturity.
This new term structure of growth expectations gives a direct measure of the financial market’s perception of economic growth for each future quarter. This measure is useful to study how certain events, such as the recent financial crisis or government policies, affect growth expectations for the years to come, across the main regions in the world economy.
In a direct comparison with other asset prices, such as nominal and real bond yields, credit spreads, and dividend yields, the newly-constructed growth expectations using equity yields fare better in forecasting future economic activity. It is well known that the forecasting relation between, for instance, nominal bond yields and future economic activity is rather unstable. This is perhaps easiest seen during periods in which interest rates are close to the zero lower-bound, where interest rates hardly move yet future growth expectations may vary substantially. Equity yields are unaffected by the zero lower-bound and therefore produce more stable forecasts of economic growth.