Debate over whether buoyant equity markets reflect durable economic gains or a widening gap between Wall Street and Main Street has intensified with the AI boom. This column argues that aggregate consumption provides a useful real-economy anchor for stock prices. Temporary departures from that anchor predict stock market returns from one quarter to two years ahead, including out of sample, but do not predict consumption growth. The evidence offers policymakers a real-time indicator of macro-financial dislocation, while stopping well short of a mechanical bubble test.
Equity markets are again confronting policymakers with an old question: when share prices rise rapidly, are they anticipating durable gains in productivity and profits, or drifting away from the real economy and fundamentals? The question is especially salient amid the AI investment boom, high market concentration, and concern that a sharp repricing could spill over to spending and financial stability. Recent VoxEU columns offer different pieces of the picture. Eeckhout (2024) attributes much of the long rise in US equity values to profits and retained earnings. Borri et al. (2026) show that markets already attach a premium to firms’ exposure to AI. For the euro area, Andersen et al. (2025) warn that equity markets may underprice downside risks to growth. Together, these contributions sharpen the policy problem: how can we tell whether movements in financial prices remain connected to macroeconomic fundamentals?
The usual answer compares stock prices with corporate cash flows, interest rates, or GDP. Our research explores another candidate: aggregate consumption. This choice follows a basic idea in asset pricing: consumption provides a broad measure of the state of the economy and may contain information about persistent economic forces that also shape firms’ future cash flows and the risks investors face. Yet the empirical link between stock market returns and consumption growth has proven to be elusive in the data. Quarterly stock returns and quarterly consumption growth are essentially uncorrelated – the well-known ‘correlation puzzle’ (Cochrane and Hansen 1992). That failure has encouraged the view that consumption is too smooth, too poorly measured, or too remote from market prices to be useful in practice.
Our central point is that the puzzle looks different when one studies levels rather than short-run growth rates. Stock prices and consumption can move very differently from one quarter to the next while still sharing a common long-run trend. In that case, weak contemporaneous correlation does not imply that markets have lost their real-economy anchor. It may instead mean that prices fluctuate substantially around it.
Measuring the gap in real time
In Favero et al. (2026), we estimate the long-run relationship between real US stock prices and real aggregate consumption. We call the temporary deviation of prices from their consumption-implied value the price–consumption cycle. A positive cycle means that prices are high relative to the level suggested by consumption; a negative cycle means that they are low.
Two features distinguish how the price–consumption cycle is identified. First, we do not impose that prices and consumption move one-for-one in the long run. The sensitivity of equity prices to consumption can differ across portfolios and change as the volatility of economic shocks changes. Second, we calculate the relationship recursively in real-time, using only the data available at each date. Future prices therefore cannot influence today’s estimate. This construction matters because full-sample estimates can create the appearance of predictability by the so called ‘look-ahead bias’, i.e. allowing information from the future to leak into the construction of the cycle to be used as a predictor of future returns.
We apply the method to the aggregate US market and to portfolios of small, big, value, and growth stocks, using quarterly data through 2024. Across all five portfolios, prices and consumption share a long-run trend. The estimated sensitivity to consumption is typically greater for small and value stocks than for big and growth stocks. It also varies over time. For the market, it stayed near to two during much of the decade after the global financial crisis and dropped sharply around the Covid-19 recession, when consumption was hit by an unusually large short-lived disturbance.
That variation is economically meaningful and time-varying. When consumption is dominated by temporary shocks, today’s consumption level becomes a noisier signal of the persistent economic forces that support asset values. A lower estimated loading in such periods need not mean that the long-run link has disappeared; it can reflect a change in the mix of permanent and transitory shocks.
What departures from the anchor predict
The price–consumption cycle mean-reverts over business-cycle horizons, generating return predictability. When prices are one percentage point above their consumption-implied value, returns over the following year are, on average, about half a percentage point lower. Conversely, prices below this anchor predict higher subsequent returns.
The predictive relation holds at horizons ranging from one quarter to two years. Averaged across the five portfolios, the cycle explains about 4% of quarterly return variation in sample, 18% of annual variation, and 32% of two-year variation. More importantly, the signal survives a recursive expanding-window forecasting exercise, delivering average out-of-sample R²s of 3.4%, 14.5%, and 21.1% at the same horizons. This performance is notable because many familiar return predictors perform well in retrospect but fail to outperform the historical-average forecast out of sample (Goyal and Welch 2008, Goyal et al. 2024). Stein and Moench (2021) similarly stress in a VoxEU column that equity-premium predictability varies over the business cycle.
The cycle is not simply another label for a standard valuation ratio. It predicts returns after controlling for price–dividend measures, the consumption–wealth ratio, and common interest-rate indicators. The cycles for the different portfolios also move together: one common component accounts for more than 80% of their variation. This points to an aggregate state variable rather than five unrelated trading signals.
Several tests sharpen the economic interpretation. First, the cycle does not forecast future consumption growth. Second, removing consumption from the long-run relationship and retaining only a time trend eliminates its return-predictive power. Our predictability results therefore cannot be attributed to generic mean reversion in prices.
Nor does it appear to arise because households sell shares to smooth temporarily depressed consumption: such a mechanism should predict a subsequent recovery in consumption, which we do not observe. The cycle does, however, contain information about dividend growth at longer horizons.
These patterns are consistent with a simple economic mechanism. Consumption contains both a persistent macroeconomic component and short-lived disturbances, while asset prices also move with time-varying required returns or beliefs. Prices can therefore depart temporarily from the consumption-related trend. As that departure unwinds, future returns become partly predictable as they revert to the common (stochastic) trend driving consumption and asset prices, even though future consumption growth does not.
A useful policy signal, not a bubble detector
The policy relevance is best understood as measurement. Central banks and financial-stability authorities routinely monitor credit spreads, leverage, valuation ratios, and market volatility. A real-time price–consumption cycle adds a different lens: it asks whether equity prices are unusually high or low relative to a slowly moving household-side fundamental. This complements evidence based on profits, cash flows, or survey expectations.
The distinction matters in episodes such as the Covid rebound or the current AI transition. A large gap between markets and current activity need not be irrational. It may reflect news about future productivity, shifts in risk-bearing capacity, or changes in discount rates. Kuvshinov and Zimmermann (2020) show in a VoxEU column that stock-market size and real activity can diverge for extended periods, even though a long-run connection remains. Our evidence adds that deviations from a consumption anchor contain systematic information about the subsequent path of returns.
That information should not be turned into an automatic trading or policy rule. The measure is estimated with uncertainty, the underlying relationships can evolve, and predictable expected returns are not synonymous with mispricing. A high reading may signal greater compensation for risk rather than a bubble. Moreover, our evidence is based on US data; its usefulness for other economies remains to be established.
The broader lesson is nevertheless clear. The apparent disconnect between stock returns and consumption growth does not imply that equity prices float free of the real economy. Consumption helps identify the slow-moving component around which prices fluctuate. Tracking departures from that anchor can deepen our understanding of macro-financial conditions – especially when market narratives move faster than the household economy beneath them.
Source : VOXeu
