Asset Premium Puzzles

Are They Really Puzzles?

Author

gitSAM

Published

March 17, 2025

Abstract
This short study revisits the Equity Premium Puzzle (EPP) and the Risk-Free Rate Puzzle (RFRP) through the lens of heterogeneous market participation and wealth inequality. We begin by outlining the standard rational-expectations framework, in which a representative-agent model generates theoretical predictions that markedly diverge from observed equity returns and risk-free rates—thereby prompting the designation “puzzle.” We then argue that these apparent anomalies are largely an artifact of oversimplified assumptions, particularly the notion that a single representative consumer meaningfully reflects the marginal investor. Drawing on empirical findings that a small subset of affluent households holds the majority of risky assets, we demonstrate how this concentrated stock market participation can reconcile theory with reality. In doing so, we suggest that, once heterogeneity in wealth and consumption is properly accounted for, these long-standing puzzles are neither inexplicable nor necessarily surprising.

1 Introduction

Since the 1980s, financial economists have grappled with two so-called “puzzles” in asset pricing theory: the Equity Premium Puzzle (EPP) and the Risk-Free Rate Puzzle (RFRP). These puzzles refer to the pronounced gap between theoretical predictions derived from standard rational-expectations models with representative-agent utility maximization and the empirical observations on asset returns and risk-free rates. In particular, Mehra and Prescott (1985) and Weil (1989) spotlighted the disparity between observed excess returns on equities and the relatively tame variability of aggregate consumption, labeling the phenomenon a “puzzle” under the canonical equilibrium framework.

Beneath both puzzles lies a single “master equation,” namely the Euler equation, which in its most general form states:

\[ \mathbb{E}_t \big[m_t R_{t+1}\big] = 1, \]

where \(m_t\) (the Stochastic Discount Factor) is closely tied to the marginal utility of consumption, and \(R_{t+1}\) represents the return on various assets. The equity premium and risk-free rate each follow from this broad formulation, yet empirical magnitudes diverge significantly from conventional theoretical predictions.

2 Theoretical Framework

2.1 Equity Premium Puzzle

The seminal work of Lucas (1978) and later Mehra and Prescott (1985) established the Equity Premium Puzzle within a representative-agent, general-equilibrium asset pricing model. The typical consumer (or representative agent) maximizes expected utility over time, commonly assumed to be CRRA (Constant Relative Risk Aversion):

\[ U(C_t) = \frac{C_t^{1-\gamma}}{1-\gamma}, \quad \gamma > 0, \,\gamma \neq 1, \]

where \(\gamma\) is the coefficient of relative risk aversion (RRA). From the consumer’s intertemporal optimization, an Euler equation emerges:

\[ \mathbb{E}_t \Big[\beta \Big(\frac{C_{t+1}}{C_t}\Big)^{-\gamma} \big(R_{e,t+1} - R_{f,t+1}\big)\Big] = 0, \]

where:

  • \(\beta\): subjective time discount factor (\(0 < \beta < 1\)),
  • \(C_{t+1} / C_t\): consumption growth rate,
  • \(R_{e,t+1}\): equity return,
  • \(R_{f,t+1}\): risk-free asset return.

Under some approximations (e.g., log-linearization of consumption growth), Mehra and Prescott (1985) famously derived:

\[ \mathbb{E}[R_e - R_f] \approx \gamma \,\sigma_c^2, \]

where \(\sigma_c^2\) is the variance of consumption growth. If the variance of consumption growth is small (on the order of 1%–2% annually), the observed 6%–8% annual equity premium can only be reconciled by positing implausibly high risk aversion coefficients (often above 20). Since typical estimates of \(\gamma\) in microeconomic or macroeconomic studies hover around 1–5, this gap forms the crux of the EPP.

2.2 Risk-Free Rate Puzzle

A closely related conundrum is the Risk-Free Rate Puzzle, initially highlighted by Weil (1989). Under the same CRRA framework and rational expectations, the Euler equation for the risk-free asset implies:

\[ 1 = \mathbb{E}_t \Big[\beta \Big(\frac{C_{t+1}}{C_t}\Big)^{-\gamma} R_{f,t+1}\Big]. \]

Approximating in logs,

\[ r_f \approx \delta + \gamma \,g_c - \frac{1}{2}\,\gamma\,(\gamma + 1)\,\sigma_c^2, \]

where:

  • \(r_f = \ln(R_f)\): the log of the risk-free rate,
  • \(\delta = -\ln(\beta)\): time preference rate,
  • \(g_c = \mathbb{E}[\ln(C_{t+1}/C_t)]\): average consumption growth rate.

Empirically, long-run real risk-free rates are typically in the 1%–3% range, whereas the above equation might predict rates of 4%–8% given plausible values for \(\gamma\), \(\delta\), and \(g_c\). Again, the severe mismatch between model forecasts and observed data has led researchers to classify it as a “puzzle.”

2.3 Critical Assumptions

To preserve tractability, the standard model assumes:

  1. A representative agent — but who truly “represents” the market?
  2. Time-separability of the utility function — ensuring that period utilities add linearly over time.
  3. Global concavity of CRRA utility — guaranteeing diminishing marginal utility at all consumption levels.

While these assumptions yield elegant closed-form solutions, they may excessively simplify real-world heterogeneity. Crucially, when the economy scales up over time, CRRA utility remains well-defined, but in practice this might obscure the role of vastly different consumption paths across distinct wealth brackets.

3 Critical Perspective

3.1 Rethinking ‘Rational Expectations’

Traditionally, Rational Expectations is seen as a condition that agents use all available information efficiently, forming unbiased forecasts. However, from a purely mathematical standpoint, “expectations” and “covariances” are simply operators for dealing with means and correlations of random variables. Such operators—particularly bilinear forms and inner products—require specific algebraic properties (linearity, symmetry, Cauchy–Schwarz inequality, etc.). While these simplifications can be powerful in physics or engineering, in economics they might be overly restrictive when applied to highly heterogeneous populations and institutions.

3.2 The Euler Equation

By construction, the Euler condition \(\mathbb{E}_t [m_{t+1} R_{t+1}] = 1\) is mathematically akin to an inner product on a probability space. This yields a focus on second moments (variance, covariance) and often leads to elliptical distribution assumptions (e.g., normal, Student-\(t\)). Real-world wealth distributions and market participation, however, may be far from elliptical in their statistical properties—especially when only a small fraction of the population holds the majority of risky assets.

3.3 Implications of Globally Concave CRRA Utility

CRRA utility, with its global concavity, implies a declining marginal utility as consumption grows. If aggregate consumption (\(C_t\)) trends upward over time, the ratio of marginal utilities \(\bigl[u'(C_{t+1}) / u'(C_t)\bigr]\) naturally declines, ensuring an inverse relationship between the SDF (\(m_{t+1}\)) and any asset (or variable) with a long-term growth trend. In equity markets, returns \(R_{t+1}\) also tend to grow over time, so \(m_t\) and \(R_{t+1}\) end up negatively correlated by construction.

If one then chooses a suitably volatile variable (with sufficient high variance) to stand in for \(m_{t+1}\), one can reconcile observed excess returns with the theoretical predictions—effectively defusing the puzzle. In that sense, the puzzle may be an artifact of incomplete modeling of real-world heterogeneity.

4 Explaining Asset Premiums

4.1 Through Market Heterogeneity

Rather than focusing exclusively on heterogeneous preferences (e.g., habit formation, behavioral biases), the perspective here is that heterogeneous economic environments—in particular, vast differences in wealth and market participation—play a decisive role:

  1. Concentrated Stock Market Participation
    • A small fraction of wealthy households hold the majority of equity. Their risk attitudes and consumption patterns have disproportionate influence on asset prices.
  2. Wealth and Consumption Inequality
    • High-net-worth individuals exhibit markedly different consumption patterns from the average household, more closely aligning with equity market fluctuations.

From this viewpoint, the high observed equity premium is not a puzzle at all once one acknowledges that a very small sub-population—namely, the extremely wealthy—holds large, volatile wealth positions that effectively determine marginal prices in the stock market. According to Federal Reserve data (Board 2025), 93% of households’ stock market wealth (though not 93% of total market capitalization) belongs to the wealthiest 10%. This implies that most aggregate equity risk and returns accrue to a relatively narrow stratum. For the “representative household,” which holds only about 7% of equity, stock price movements have minimal impact on its marginal utility of consumption. Hence, standard representative-agent formulations fail to capture the truly relevant marginal investor.

Empirical evidence from the following studies supports this line of reasoning:

  • Basak and Cuoco (1998): Demonstrates that limited stock market participation elevates risk premia and depresses risk-free rates.
  • Yogo (2006): Shows that consumption by wealthy households closely tracks equity returns, reinforcing the link between high-end consumption dynamics and asset prices.
  • Gomes and Michaelides (2008): Connects growing income inequality with stock market participation patterns, resulting in rising equity premia.
  • Lettau, Ludvigson, and Ma (2019): Shows that a single macroeconomic factor tied to capital share (reflecting wealthy shareholders’ consumption) can explain a broad range of cross-sectional stock return premia.

4.2 Limitations

Despite offering a plausible explanation, this heterogeneity-based view faces practical hurdles:

  • Low frequency data: Due to infrequent reporting on high-net-worth wealth, applying short-term no-arbitrage principles in cross-sectional asset pricing is problematic.
  • Difficulty of reconciling EMH: Efficient Markets Hypothesis (EMH) posits that arbitrage opportunities vanish quickly, but wealth data often lack the granularity or frequency to confirm this.
  • Long-run identification: Changes in upper-tier wealth or consumption may be valid for a long-run SDF (\(m\)), yet verifying this for short-run asset pricing remains challenging.

5 Conclusion

The Equity Premium Puzzle and Risk-Free Rate Puzzle have dominated discussions in asset pricing for decades. However, labeling them as genuine “puzzles” may reflect an artifact of restrictive models that hinge on a single representative agent, uniform preferences, and high-level assumptions about consumption growth. By introducing heterogeneous market participation, particularly the reality that a small fraction of wealthy agents holds the lion’s share of risky assets, one finds that what appears to be a puzzle for the average consumer is, in fact, quite explicable among those who actually drive stock prices.

In short, when empirical ownership and wealth concentration data are properly accounted for, the puzzling gaps between theory and observation can diminish or disappear. The challenge remains to integrate heterogeneous agent frameworks with accurate micro-level data on wealth and consumption in order to provide a more comprehensive understanding of asset prices—an endeavor that holds promise for reconciling the so-called “puzzles” with empirical reality.

References

Basak, Suleyman, and Domenico Cuoco. 1998. “An Equilibrium Model with Restricted Stock Market Participation.” The Review of Financial Studies 11 (2): 309–41.
Board, Federal Reserve. 2025. “Distributional Financial Accounts (DFAs).” https://www.federalreserve.gov/releases/z1/dataviz/dfa/.
Gomes, Francisco, and Alexander Michaelides. 2008. “Asset Pricing with Limited Risk Sharing and Heterogeneous Agents.” The Review of Financial Studies 21 (1): 415–48.
Lettau, Martin, Sydney C Ludvigson, and Sai Ma. 2019. “Capital Share Risk in US Asset Pricing.” The Journal of Finance 74 (4): 1753–92.
Yogo, Motohiro. 2006. “A Consumption-Based Explanation of Expected Stock Returns.” The Journal of Finance 61 (2): 539–80.