Can Demography Help Us Forecast Stock Returns?

One of the most debated issues in financial econometrics is the empirical relevance of the dividend-price (DP) ratio for forecasting the long-horizon stock market. Unlike the dismay short-run prediction where the robustness is frequently questioned by researchers (notably, Goyal and Welch (2008)), the idea of the long-run predictive ability seems to be appealing and well supported by dynamic dividend growth theory. Yet, the DP ratio’s high persistence contradicts the fundamental stationarity hypothesis required by the theory and greatly limits its predictive ability empirically (known as the Stambaugh (1999) bias). In other words, if we can find a way to reduce this persistence, we may accomplish much better long-horizon forecasting for stock returns.

Recent work by Favero et al.(2011) (hereafter, FGT) addressed this issue. They find the slowly evolving demographic trends, especially the middle to young ratio (MY), strongly associated with the valuation. The idea behind it is simple but constructive. As people are more likely to borrow when they are young but save when they are middle-aged, a high (low) MY ratio implies an excess demand for saving (consumption), which surges (tumbles) the equilibrium prices of financial assets and thus the DP ratio. Therefore, ideally, once we remove the demographical component trend from the DP ratio, the newly adjusted DP ratio should exhibit less persistence and improve the long-run stock return forecasts.

FGT test their conjecture by proposing a predictive regression model including the lagged MY ratio as a predictor. They run a “horse race” among model candidates, and the results are pretty striking – MY ratio significantly improves the prediction under both in-sample and out-of-sample analysis. 

Henceforth, more empirical evidence appears to support the low-frequency co-movement between the demographic ratio and the DP ratio and its usefulness in long-horizon forecasts. Among others, Chen et al. (2021) (hereafter CGMP) show that we can further extend the predictive ability of the DP ratio by absorbing some slowly evolving information in the future. By investigating the data, CGMP finds a close match between the dynamics of the demographic variables and the common factors of stock returns (see figure below). CGMP also notice a unique property – the demographic ratios are much more likely to be predicted precisely than DP ratios. In conventional prediction, including FGT, we can only use real-time information to forecast the future. However, if we can somehow reasonably forecast the demographic ratio and we admit there is a strong link from demographical structure to the DP ratio and the stock return, why not use the demographic projections (for example from the Census Bureau) to project the long-run stock return?

Common factor in stock returns and valuation ratios, and demographics

Source: Chen et al., 2021.

This approach enables us to provide an update of FGT’s real-time unconditional stock return forecasts. Moreover, as the demographic projection is essentially a source of another forecast, this approach also falls into a conditional forecast category, which is growing rich in literature and quite popular among researchers and policymakers. For example, although we may not be able to predict the stock return unconditionally, we may forecast future stock return conditional on different future demographic scenarios (say a lower/higher birth incentive or a more strict/looser immigration policy or a deadly pandemic like Covid-19 occurs).

To examine this idea, CGMP propose a new predictive regression model replacing the lagged MY ratio of FGT with the MY ratio projection. Due to the sparse projection data, CGMP first use the true future MY ratio as the demographic “projection” and conducts the conditional forecast. The baseline analysis confirms the previous evidence that demographic ratios improve the long-horizon forecasts. They also find that their proposed model outperforms FGT uniformly in the out-of-sample study.

After observing the promising results, CGMP collect all available projections from Census Bureau and build a large real-time demographic projection dataset. With the actual projection data, CGMP examines the model’s unconditional forecasting ability. Again, the proposed model mostly outperforms FGT, although the gap between the two methods narrows down in some cases.

Of course, the results mentioned above may be doubted by the sample choice, the validity of the statistical test, and other notorious but ubiquitous problems in the game of the cherry-picking competition. Sadly, we cannot verify those questions until the very far end of the future. Knowledge of prediction is always like navigating in an ocean of uncertainties through the isle of certainties. The same is true for stock returns. With the tool of demographic ratio projection, the sea is getting smaller. But be careful. The navigation is also getting blurred.

Chaoyi Chen

Chaoyi Chen joined the MNB and the MNB Institute in 2020. He received a PhD in Economics from University of Guelph in 2019. His research interests are econometrics and applied econometrics. His research topics have included the threshold regressions,  long-horizon regressions, nonstationary time series regressions, and applications on macroeconomics.


References:

Chen, C., N. Gospodinov, A. Maynard, and E. Pesavento (2021), Long-Horizon Stock Valuation and Return Forecasts Conditional on Demographic Projections, mimeo.

Favero, C. A., Gozluklu, A. E. and Tamoni, A. (2011), Demographic trends, the dividend-price ratio, and the predictability of long-run stock market returns, Journal of Financial and Quantitative Analysis, 46(05), 1493-1520.

Goyal, A. and Welch, I. (2008), A comprehensive look at the empirical performance of equity premium prediction, Review of Financial Studies, 48(2), 663-670. Stambaugh, R. F. (1999), Predictive regressions, Journal of Financial Economics, 54, 375-421.


Főoldali kép forrása: pixabay.com