Asset pricing with long-run disaster risk

Rujie Fan; Chao Xiao

doi:10.1371/journal.pone.0287687

Abstract

Traditional disaster models with time-varying disaster risk are not perfect in explaining asset returns. We redefine rare economic disasters and develop a novel disaster model with long-run disaster risk to match the asset return moments observed in the U.S. data. The difference from traditional disaster models is that our model contains the long-run disaster risk by treating the long-run ingredient of consumption growth as a function of time-varying disaster probability. Our model matches the U.S. data better than the traditional disaster model with time-varying disaster risk. This study uncovers an additional channel through which disaster risk affects asset returns and bridges the gap between long-run risk models and rare disaster models.

Citation: Fan R, Xiao C (2023) Asset pricing with long-run disaster risk. PLoS ONE 18(6): e0287687. https://doi.org/10.1371/journal.pone.0287687

Editor: Nicolò Pecora, Università Cattolica del Sacro Cuore Sede di Piacenza e Cremona Facoltà di Economia: Universita Cattolica del Sacro Cuore Facolta di Economia e Giurisprudenza, ITALY

Received: February 27, 2023; Accepted: June 9, 2023; Published: June 27, 2023

Copyright: © 2023 Fan, Xiao. This is an open access article distributed under the terms of the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited.

Data Availability: The main data are from Robert J. Barro's website (https://scholar.harvard.edu/barro/data_sets; please see "Barro-Ursúa Macroeconomic Data (2010)"), World Development Indicators (https://databank.worldbank.org/source/world-development-indicators) and Robert Shiller's website (http://www.econ.yale.edu/∼shiller/data.htm; please see "U.S. Stock Markets 1871-Present and CAPE Ratio"). Other data are within the paper and its Supporting Information file.

Funding: The author(s) received no specific funding for this work.

Competing interests: The authors have declared that no competing interests exist.

Introduction

Traditional rare disaster models play an essential role in explaining asset returns [1–10]. At present, there are three main methods for modeling disaster risk: (i) the static disaster probability (a constant) [2, 4, 6, 11]; (ii) the exogenous time-varying disaster probability (e.g., a square-root process) [8, 12–15]; (iii) the endogenous time-varying disaster probability (e.g., a Bayesian learning process) [16–24]. These studies have shown that the time-varying disaster risk, rather than the static disaster risk, is the key to successfully interpreting asset returns by traditional disaster models. However, traditional disaster models with time-varying disaster risk are far from perfect in explaining asset returns.

In this paper, we construct a new discrete-time disaster model to explain asset returns in the United States. We assume that the long-run ingredient in consumption growth rates is a function of the time-varying disaster probability. Consumption growth faces the standard diffusion risk, the time-varying disaster risk, and the long-run disaster risk; the latter two are two manifestations of disaster risk. A disaster is modeled as a jump with a negative size. Meanwhile, we assume that the jump follows a Bernoulli distribution, the jump size follows a gamma distribution, and time-varying disaster probability obeys a square-root process [25]. Using the technique of linearity-generating processes [26], our model is tractable, and all asset prices are solved in closed form.

The solution for our model reveals that the time-varying disaster risk alone is insufficient to generate the equity premium, the volatility of equity returns, and the Sharpe ratio observed from the U.S. data. The model fits well with the data by combining time-varying disaster risk with long-run disaster risk. Adding the long-run disaster risk to the traditional time-varying disaster framework not only comprehensively improves the model’s ability to explain asset returns but also closely links rare disaster models with long-run risk models. Moreover, compared to the traditional time-varying disaster model, our model more strongly implicates equity premiums’ predictability and consumption growth’s unpredictability. All traditional disaster pricing models imply that consumption growth is predictable in long horizons. However, consumption growth is unpredictable in the data, especially in long horizons. The long-run disaster risk can mitigate such conflicts because it has a much smaller impact on consumption growth than on the price-dividend ratio.

Few studies combine the long-run risk with the disaster risk to explain asset returns. Bansal et al. [27] propose a temperature-augmented long-run risk model in which the long-run ingredient in consumption growth rates is subject to standard stochastic shocks and natural disasters caused by climate change. In contrast, we assume that the long-run ingredient is a linear function of time-varying disaster probability following a square-root process [25]. Barro and Jin [28] build a model with rare disasters and long-run risk and assume that the probability of disasters is constant and that the long-run ingredient in consumption growth rates is consistent with that in Bansal and Yaron [29]. Their assumption lacks strong links between traditional disaster risk and long-run risk. In our model, the dynamics of both the jumps (disasters) and the long-run ingredient depend on time-varying disaster probability. There is also some other literature related to the assumptions in the theoretical model of this paper. According to Barro and Jin [30], the distribution of transformed sizes of disasters, as defined by Barro and Ursúa [2], fits well with a power-law density. However, according to our definition of disasters, the disaster sizes tend to follow a gamma distribution. Similar to this paper, Bansal et al. [27] and Marfè and Pénasse [12] also assume that the disaster size follows a gamma distribution. To explain aggregate stock market volatility, unlike the existing literature on models with Bayesian learning or a Markov process [3, 16, 24, 30, 31], Wachter [8] assumes that the time-varying disaster probability obeys a square-root process [25]. Wachter [8] does not introduce long-run risk in the theoretical model; while its model can generate higher stock return volatility than dividend volatility, it also leads to excessive volatility in consumption and dividends.

Rare economic disasters

As in Marfè and Pénasse [12], we define a rare economic disaster as a severe decline in consumption during a given year. A country experiences a rare economic disaster in the year t + 1 if Δc_t+1 falls by two times the standard deviation (SD) from its growth path: (1) where Δc_t+1 = log C_t+1 ‒ log C_t, and N_t+1 = 1 means an economic disaster with a disaster size of mean (Δc) ‒ Δc_t+1 occurred in period t + 1.

We use long-term real per capita consumption expenditure data from 42 countries to assess historical economic disasters. Among them, the data on real consumption expenditure per capita before 2006 is provided by Barro and Ursúa [2]. We refer to Barro and Ursúa [2] to expand the data to 2021 and calculate the consumption growth rate from the consumption data. Descriptive statistics on the growth rate of real consumption expenditure per capita by country and the distribution of rare economic disasters on the timeline are included in detail in S1 Data. The sample size of the growth rate of real consumption expenditure per capita is equal to 5348. Table 1 presents descriptive statistics of rare economic disasters. The number of disasters is 183, the probability of a disaster is 0.0342, the average size of disasters is 0.1827, and the standard deviation of disaster sizes is 0.0907.

Download:

Table 1. Descriptive statistics of rare economic disasters.

https://doi.org/10.1371/journal.pone.0287687.t001

Next, we continue to study the distribution characteristics of disaster sizes. Barro and Jin [30] argue that the distribution of transformed sizes of disasters fits well with a power-law density. However, as shown in Fig 1, the Gamma density rather than the power-law density better fits the size distribution of our disasters. Fig 2 shows the distribution of disaster probabilities close to the normal distribution.

Download:

Fig 1. Distribution of disaster sizes.

https://doi.org/10.1371/journal.pone.0287687.g001

Download:

Fig 2. Distribution of disaster probability.

https://doi.org/10.1371/journal.pone.0287687.g002

The model

Agents have a recursive but not time-separable preference [32, 33]: (2) where variables U_t and C_t represent the utility and consumption of agents, δ is the subjective discount factor, and ψ and γ measure the elasticity of intertemporal substitution and the relative risk aversion, respectively. Eq (2) suggests that agents are risk-averse about potential future consumption risk. In this paper, we refer to Wachter [8] and Vissing-Jørgensen [18] to set ψ in Eq (2) to one for tractability. It is worth noting that the reasonable value of ψ is controversial [8, 9, 18, 29, 34–38].

Proposition 1

When ψ is equal to one, the agents’ preference is as follows:

(3)

The proof of proposition 1 is shown in Appendix A in S1 Appendix. Agents receive an endowment consumption stream C_t. The log consumption growth rate is given by (4) where Δc_t+1 = log C_t+1 ‒ log C_t,μ is the constant term, σ represents the volatility of consumption growth without disaster risk, and ε_t+1 ∼ N(0,1). N_t+1 follows a Bernoulli distribution: (5)

In Eq (5), p_t represents the disaster probability. p_t obeys a square-root process [25]: (6) where 1 ≥ p_t ≥ 0,θ is the constant term, ϕ represents the rate of mean reversion of disaster probability, σ_p is the volatility parameter of disaster probability, and ε_t+1 ∼ N(0,1). The unconditional mean of disaster probability is θ/(1‒ϕ), and its unconditional variance can be written as . Therefore, τp_t in Eq (4) means that the disaster probability has a long-run negative impact on consumption growth (τ < 0). In this paper, the AR(1) process of the long-run component of consumption growth is as follows: (7)

Eq (4) can be rewritten as: (8)

In Eq (4), ϑ_t+1 indicates the disaster size. We assume that it follows a gamma distribution Γ(λ,ι) with a moment-generating function given by (9) where the mean and variance are equal to λ/ι and λ/ι², respectively.

Asset return

Stochastic discount factor

Eq (3) implies that the log stochastic discount factor is (10) where ζ_t+1 = log U_t+1 –log C_t. An affine expression for the log stochastic discount factor can be obtained by the following assumptions and propositions. We first give assumption 1.

Assumption 1.

ζ_t+1 = ϖ_t+1 – ϑ_t+1N_t+1, where ϖ_t+1 follows a normal distribution and is independent of ϑ_t+1 N_t+1, E_t(ϖ_{t+1) =} v_t, and .

Proposition 2.

According to Assumption 1, the following equation can be derived: (11)

The proof of proposition 2 is shown in Appendix B in S1 Appendix. Moreover, we give propositions 3 and 4.

Proposition 3.

If we define y_t = log U_t−log C_t, then Eq (3) can be transformed into (12)

Proposition 4.

y_t = L₀ + Lp_t, where , . Meanwhile, we can write that v_t = L₀ + μ + Lθ + (Lϕ+ τ)p_t and .

Proofs for propositions 3 and 4 are shown in Appendices C and D, respectively. Therefore, Eq (10) can be rewritten as follows: (13)

The risk-free rate

The log risk-free rate satisfies (14)

In the standard model without disaster risk, the log risk-free rate is expressed by the term above the first bracket in Eq (14); μ refers to the average growth rate of consumption without disaster risk; −log δ represents the rate of time preference; (1–2γ)σ² / 2 stands for the effect of precautionary saving due to the standard risk (σε_t+1) on the log risk-free rate. The term above the second bracket in Eq (14) is derived from disaster risk. In theory, because people are risk-averse, an increase in the probability of disasters will strengthen people’s willingness to save, thereby reducing the risk-free rate. Thus, if we define that should be negative. As shown in Fig 3, f (γ) decreases as risk aversion increases.

Download:

Fig 3. Relative risk aversion and disaster risk.

https://doi.org/10.1371/journal.pone.0287687.g003

When is equal to zero, Eq (14) reflects the relationship between the risk-free rate and the disaster probability in traditional time-varying disaster risk models. As shown in Eq (8), we introduce the long-run disaster risk into consumption growth by setting τ not equal to zero. Therefore, both time-varying and long-run disaster risk are included in our model. Fig 4 shows that the risk-free rate is a decreasing function of disaster probability. The dashed line represents the traditional time-varying disaster risk model, while the solid line represents our model. When the disaster probability equals zero, the risk-free rates derived from the two models are the same. The solid line lies below the dashed line, as long-run disaster risk further boosts savings. The difference between the solid and dashed lines reflects the effect of long-run disaster risk on the risk-free rate. This effect increases significantly as the probability of disasters increases. Fig 5 indicates that the long-run disaster risk can increase the volatility of the risk-free rate.

Download:

Fig 4. Effect of the long-run disaster risk on the risk-free rate.

https://doi.org/10.1371/journal.pone.0287687.g004

Download:

Fig 5. Effect of the long-run disaster risk on the volatility of the risk-free rate.

https://doi.org/10.1371/journal.pone.0287687.g005

The equity return

Let Δd_t+1 denote the dividend growth rate. We model the dividend growth rate as a function of the consumption growth rate [8, 35, 39–41]: (15) where Φ is the leverage parameter. The equity return, r_e,t+1, can be expressed in terms of the log price-dividend ratio (pd_t+1 and pd_t) as (16)

First, linearizing Eq (16) around the steady state of pd_t defined as pd_ss, we obtain: (17) where . According to proposition 5 below, we can obtain the affine expression of r_e,t+1.

Proposition 5.

(18)

where , .

The proof of proposition 5 is shown in Appendix E in S1 Appendix. Fig 6 illustrates how to solve a fixed point problem to find the steady state of log price-dividend ratios under the underlying parameters. We have found pd_ss when its assumed value equals its unconditional mean (E(pd_t)).

Download:

Fig 6. The steady state of the log price-dividend ratio.

https://doi.org/10.1371/journal.pone.0287687.g006

Fig 7 presents the effect of disaster probability on the log price-dividend ratio in different disaster models. Although disaster risk is modeled differently, the log price-dividend ratio is always a monotonically decreasing function of disaster probability. The dashed line is below the dotted line. It is steeper than the dotted line, indicating that substituting time-varying disaster risk for static disaster risk in theoretical modeling enables the model to generate a lower log price-dividend ratio that is more sensitive to disaster probability. The solid line is below the dashed line. It is steeper than the dashed line, indicating that incorporating long-run disaster risk into the traditional time-varying disaster risk model can further reduce the log price-dividend ratio and enhance the sensitivity of the log price-dividend ratio to disaster probability.

Download:

Fig 7. Influence of disaster probability on log price-dividend ratio in different disaster models.

https://doi.org/10.1371/journal.pone.0287687.g007

The corresponding relationship between equity premium and disaster probability in different disaster models is shown in Fig 8. When the disaster probability is zero, all three types of models are simplified to the standard diffusion model without disaster risk. Thus, three straight lines have an intersection. The risk premium represented by this intersection is derived from standard diffusion risk. It is rational in economics that the return people require for holding equity increases with the probability of disasters. However, in different disaster models, there are significant differences in the equity premium caused by the same disaster probability. Among them, the equity premium of the static disaster risk model (dotted line) is the smallest, the equity premium of the time-varying disaster risk model (dashed line) is second, and the equity premium of the time-varying disaster risk model, including long-run disaster risk (solid line), is the largest. The slope of the dashed line is greater than that of the dotted line, and the slope of the solid line is greater than that of the dashed line. The difference between the solid and dashed lines represents the equity premium due to long-run disaster risk and shows that it is the main component of the total risk premium.

Download:

Fig 8. Disaster probability and the equity premium.

https://doi.org/10.1371/journal.pone.0287687.g008

How does equity volatility relate to disaster probability in our model? Fig 9 shows that equity volatility is a concave and increasing function of the disaster probability. When the probability of disasters is close to zero, the volatility of equity returns for all three models is close to that of the dividend in non-disaster times (Φσ). Long-run disaster risk not only significantly increases equity premiums but also significantly increases the volatility of equity returns. That is, an increase in the equity premium is accompanied by an increase in the volatility of equity returns. It is worth noting that equity premium increases linearly with the disaster probability, while the volatility of equity returns increases with a square root.

Download:

Fig 9. Disaster probability and volatility of equity returns.

https://doi.org/10.1371/journal.pone.0287687.g009

The Sharpe ratio equals the equity premium divided by the volatility of equity returns and measures the premium per unit of risk. The impact of disaster probability on the Sharpe ratio is shown in Fig 10.

Download:

Fig 10. Disaster probability and the Sharpe ratio.

https://doi.org/10.1371/journal.pone.0287687.g010

Calibration and simulation

Since we have analyzed in detail how the three types of disaster models differ in understanding asset returns, we next focus on matching these models to real data. Our model measures time in years and the parameter values are supplied by this.

Calibration

Table 1 has presented the descriptive statistics of rare economic disasters. In our model, the constant term in the disaster probability process, θ, determines the probability of disasters when ϕ is fixed; the larger θ, the higher the average probability of disasters. Because of E(p_t) = θ/(1-ϕ), the parameter θ can be calculated. Also, since the disaster size is assumed to obey a gamma distribution, as shown in Eq (9), we can deduce λ and l under the premise of knowing the mean and standard deviation of disaster sizes. Part A of Table 2 gives the values of disaster parameters (λ, l, and θ). In Part B of Table 2, we calibrate the standard parameters (δ, γ, ϕ, and Φ) with reference to Wachter (2013). From Eq (4), it can be seen that the volatility of consumption growth during the non-disaster period in this paper is not only affected by the short-run shock (σε_t+1) but also by time-varying disaster probability (p_t). Therefore, we set σ to a lower value than the value of this parameter in the traditional time-varying disaster risk model (0.02).27.36x10-4

Download:

Table 2. Parameters for simulation.

https://doi.org/10.1371/journal.pone.0287687.t002

Simulation results

Table 3 shows moments from the simulation data of these models and moments from the long historical time series (1870–2006) of the United States. Column 2 in Table 3 shows the target values of asset-pricing statistics. These targets are the mean and standard deviation of consumption growth, average equity premium, the mean and standard deviation of equity returns, the standard deviation of dividend growth, the Sharpe Ratio, and the mean and standard deviation of the risk-free rate. Column 3 in Table 3 refers to the static disaster risk model, in which the disaster probability is constant (i.e., σ_p = 0). The model does not include long-run disaster risk, so we set τ equal to zero. Given the calibrated parameters in Table 2, the model turns out to require a mean of consumption growth in non-disaster times, μ, of 0.025 to fit the mean of consumption growth (1.85%). The results show that the static disaster risk model seriously underestimates the equity premium, the volatility of equity returns, and the volatility of the risk-free rate. It is implausible that the volatility of the risk-free rate is equal to zero. The standard deviation of equity returns is equal to the standard deviation of dividend growth, which is also inconsistent with empirical evidence. Thus, the static disaster risk model is insufficient to characterize asset returns. Extensive literature shows that time-varying disaster risk models outperform static disaster models in understanding asset returns. Column 4 in Table 3 shows results with the model that includes only time-varying disaster risk. So, σ_p is set to be greater than zero, and τ is set to zero. Specifically, σ_p is set to 0.067 with reference to Wachter [8]. The value of μ in column 4 equals that in column 3. Overall, replacing the static disaster risk (Column 3) with the time-varying disaster risk (Column 4) generates moderate improvements in the results.

Download:

Table 3. Moments (annual).

https://doi.org/10.1371/journal.pone.0287687.t003

Moments for a model with time-varying and long-run disaster risk are given in column 5 of Table 3. The parameter σ_p in column 5 is set to 0.067 to facilitate a comparison of the results in columns 5 and 4. It can be seen from Eq (4) that τ ≠ 0 means that consumption growth contains a long-run ingredient and that τ < 0 means that the negative impact of the long-run ingredient on consumption growth increases with the increase of disaster probability. Then, a key question is what value should τ take? From historical documents, it is difficult for us to find an answer to this question. Next, we try to determine the value of τ by matching it with the classic long-run risk model. In the classic long-run risk models, the long-run component (z_t) is modeled as highly persistent AR(1) processes: (19) where ε_z,t+1 ∼ N(0,1), and E(z_t) = 0. Colacito and Croce [38] take the values of ρ and σ_z in Eq (19) to be 0.985 and 2.62 × 10⁻³, respectively. The unconditional variance of z_t is equal to 2.30 × 10⁻⁴. Based on Eq (7), an expression for the unconditional variance of the long-run component in this paper can be written: . Assuming and knowing θ, σ_p and ϕ, τ = −0.48 can be deduced. We set the value of τ in column 5 to -0.13 to prevent the model from generating an excessively high equity premium. Unlike the long-run ingredient in classic literature [29, 38, 40], which has no clear economic connotation, the long-run ingredient in this paper is a linear function of disaster probability (see Eq (7)). The μ in column 5 is larger than in column 4 to offset the effect of τp_t on consumption growth. The results in column 5 of Table 3 suggest that long-run disaster risk plays an important role in explaining asset returns. Comparing column 5 and column 4 in Table 3, it can be seen that the inclusion of the long-run disaster risk in traditional time-varying disaster risk models can bring a higher equity premium, a higher equity return, a higher equity return volatility, a lower Sharpe ratio, and a higher risk-free rate volatility. The above comparison shows that not only are long-run risk and disaster risk complementary in explaining asset returns but they may be driven by common factors such as disaster probability. Although Barro and Jin [28] also point out that rare disaster models and long-run risk models are complementary, the disaster probability is static in their model, and the long-run risk still lacks a connotation. In Barro and Jin [28], disaster risk and long-run risk affect asset returns independently in their ways.

It can be seen from the above that the consumption-based capital asset pricing model established in this paper can well explain asset returns. So how do rare economic disasters affect asset returns in this article? In theory, a consumption-based capital asset pricing model might perfectly explain asset returns only if all risks of consumption growth are priced in. According to the existing literature, risks to consumption growth include short-run risks and long-run risks. In this paper, rare economic disasters affect consumption growth through two channels: first, random rare economic disasters are one of the sources of short-run risks for consumption growth; secondly, the probability of rare economic disasters with high first-order autocorrelation is a source of long-run risks for consumption growth. Eq (4) shows that when no rare economic disaster occurs in period t + 1, the expected growth rate of consumption in this paper is μ + τp_t, which is a constant in the traditional model. In practice, even if there is no rare economic disaster in the future, people’s expected consumption growth rate will change with the probability of a rare economic disaster in the future. Furthermore, the agent has recursive preferences in our model, and the agent’s need to address risk ahead of time leads to the pricing of expected short-run and long-run risks.

Implied disaster probability

We combine the S&P 500 price-earnings ratio with the models in this subsection to derive the implied disaster probability. Eq (18) shows that the price-dividend ratio is a monotonically decreasing function of disaster probability (A < 0). Therefore, the implied disaster probability can be calculated if the price-dividend ratio is known. Wachter [8] points out that for long-time series data, the price-earnings ratio from the data should be used to match the price-dividend ratio of the theoretical model. Robert J. Barro’s website (https://scholar.harvard.edu/barro/data_sets) provides the S&P 500 price-earnings ratio from 1880 to 2021 and the method for calculating the price-earnings ratio. When calculating the implied disaster probability, we set the mean-removed price-earnings ratio in the data equal to the mean-removed price-dividend ratio in the model. When the disaster probability is negative, it is set to zero.

It can be seen from Table 1 that the average probability of rare economic disasters is 0.0342. According to the calibrated parameters (θ, ϕ, and σ_p), we implicitly assume that the standard deviation of disaster probability is 0.0316. Fig 11 depicts the implied probability of disasters for the United States calculated using P/E data. For the time-varying disaster model (dashed line), the mean and standard deviation of implied disaster probability are 0.2189 and 0.0943, respectively; for our model with time-varying and long-run disaster risk (solid line), they are 0.0607 and 0.0436, respectively. The solid line in Fig 11 better reflects the historical real disaster probability. These results also indirectly support our assumption that the long-run ingredient in consumption growth rates is a function of disaster probability.

Download:

Fig 11. Implied disaster probability in the United States.

The horizontal dashed lines represent the means of the two curves, respectively.

https://doi.org/10.1371/journal.pone.0287687.g011

Predictability

Bansal and Yaron [29] argue that there is a small but persistent component of U.S. consumption growth over the past century, which they call long-run risk. The long-run risk in consumption growth rates does not conflict with empirical evidence that consumption growth is close to a random walk. Long-run risk, while only explaining a small part of consumption volatility, explains a large part of asset returns and their volatility under recursive preference. Estimating long-run risk in consumption growth rates is a big challenge. Colacito and Croce [40] refer to Bansal et al. [36], using risk-free rates, consumption-output ratios, lagged consumption growth, lagged price-dividend ratios, and default premium to systematically estimate the long-run ingredient of consumption growth; the results show that the long-run ingredient does exist.

The ability of disaster probability to predict macroeconomic growth is a common feature of various disaster models. Although this feature contradicts the unpredictability of consumption growth pointed out by empirical evidence, the importance of disaster models for understanding asset returns is unquestionable. Therefore, we need a rare disaster model that can mitigate the above conflicts efficiently. Most existing disaster models lack modeling of the long-run ingredient of economic growth. In our model, we assume that the long-run ingredient in consumption growth rates is a linear function of the disaster probability. This is an innovation. Next, we will prove that our assumption about the consumption process is closer to the dynamics of actual consumption data than those in traditional time-varying disaster risk models. That is, the predictability of consumption growth by disaster probability is significantly reduced in our model.

In reality, the probability of disaster is unobservable. Eq (18) shows that the log price-dividend varies linearly with the probability of a disaster. Therefore, we can use the log price-dividend ratio instead of disaster probability to predict consumption growth. Long-horizon consumption growth is regressed on the lagged log price-dividend ratio in simulated data and actual data for the United States from 1880 to 2021. The regression equation is as follows: (20)

As shown in Panel A of Table 4, in U.S. data, while long-horizon consumption growth is unpredictable, short-horizon consumption growth is predictable to a certain extent. At the one-year horizon, β₁ is 0.01, R² is 0.03, and T-statistic is equal to 2.11; otherwise, the regression coefficients are insignificant, and the maximum R² does not exceed 0.01. In Panel B of Table 4, consumption growth is regressed on the lagged log price-dividend ratio in data simulated from the time-varying disaster risk model. In Panel C of Table 4, the results refer to our model. Unsurprisingly, in both models, the price-dividend ratio significantly predicts consumption growth. In each horizon, the R² in Panel C is close to that in Panel B, and the T-statistic in Panel C is close to that in Panel B. However, the β₁ in Panel C is significantly smaller than in Panel B on each horizon. Obviously, in terms of unpredictability in consumption growth rates, the data simulated from our model is more consistent with the actual data. This is because the long-run disaster risk has a much smaller impact on consumption growth than on the log price-dividend ratio, reducing consumption growth’s predictability.

Download:

Table 4. Long-horizon regressions: Consumption growth.

https://doi.org/10.1371/journal.pone.0287687.t004

How does long-run disaster risk in this paper affect the predictability of equity premiums? The long-horizon equity premium is regressed on the lagged log price-dividend ratio in simulated data and actual data for the United States from 1991 to 2021. The regression equation is as follows: (21)

Panel A of Table 5 shows that the regression coefficient on the log price-dividend ratio (β₁) is significantly less than zero: a high disaster probability corresponds to a low log price-dividend ratio (see Eq (18)) and thus also predicts a high future expected equity premium. β₁ is -0.13 at the one-year horizon, decreasing to -0.86 at the ten-year horizon. Meanwhile, R² is 0.09 at the one-year horizon, rising to 0.43 at the ten-year horizon. Panel B reports the moments of the model with time-varying disaster risk. In panel B, the regression coefficients are insignificant at both the one- and two-year horizons; this suggests that the log price-dividend ratio cannot predict the short-horizon equity premium. Panel C reports the results of our model with time-varying and long-run disaster risk. In panel C, the regression coefficient is significant on each horizon; β₁ is -0.12 at the one-year horizon, decreasing to -0.75 at the ten-year horizon; R² is 0.05 at the one-year horizon, rising to 0.22 at the ten-year horizon. The results in plane C are closer to those in plane A than those in plane B because significantly reduce the price-dividend ratio. Moreover, β₁ in plane C divided by β₁ in plane B equals 2.4 at the one-year horizon, decreasing to 1.15 at the ten-year horizon; this is because the long-run disaster risk not only increases the total amount of risk in the economy but also changes the distribution of equity premiums over time. With recursive preference, agents are risk-averse to future risk; with the extension of the horizon, the impact of the long-run disaster risk on equity premiums gradually weakens.

Download:

Table 5. Long-horizon regressions: Equity premium.

https://doi.org/10.1371/journal.pone.0287687.t005

Conclusions

To improve traditional rare disaster models, we redefine macroeconomic disasters and propose a novel rare disaster model with time-varying and long-run disaster risk. Using the technique of linearity-generating processes, the model is tractable, and all asset prices are solved in closed form. Our main conclusions are as follows.

Firstly, incorporating long-run disaster risk into traditional rare disaster models can increase the volatility of asset returns and the sensitivity of asset returns to disaster risk. Secondly, our model with time-varying and long-run disaster risk matches the U.S. data better than traditional rare disaster models with only static or time-varying disaster risk. In our model, disaster risk affects asset returns not only through a short-run channel (the time-varying disaster risk) but also through a long-run channel (the long-run disaster risk).

Our model not only facilitates the analysis of the mechanism by which disaster risk affects asset returns but also provides a new potential channel for disaster risk to affect asset returns. It also enriches the connotation of long-run risk and strengthens the inherent connection between long-run risk models [29] and traditional rare disaster models [8]. Since the disaster probability is unobservable, it is unrealistic to directly prove our model’s fundamental assumption that the long-run ingredient of consumption growth is a function of time-varying disaster probability. Providing more evidence for this assumption may be a future research direction. Extending our theoretical framework to pricing other asset classes (e.g., bonds, exchange rates, and options) is also a future research direction.

Supporting information

S1 Data.

https://doi.org/10.1371/journal.pone.0287687.s001

(XLSX)

S1 Appendix.

https://doi.org/10.1371/journal.pone.0287687.s002

(DOCX)

References

1. Barro RJ. Rare disasters and asset markets in the twentieth century. The Quarterly Journal of Economics. 2006; 121(3): 823–866. https://doi.org/10.1162/qjec.121.3.823.
- View Article
- Google Scholar
2. Barro RJ, Ursúa JF. Macroeconomic crises since 1870. National Bureau of Economic Research. 2008. https://doi.org/10.3386/w13940.
- View Article
- Google Scholar
3. Gourio F. Disasters and recoveries. American Economic Review. 2008; 98(2): 68–73. https://doi.org/10.1257/aer.98.2.68.
- View Article
- Google Scholar
4. Barro RJ. Rare disasters, asset prices, and welfare costs. American Economic Review. 2009; 99 (1): 243–64. https://doi.org/10.1257/aer.99.1.243.
- View Article
- Google Scholar
5. Julliard C, Anisha G. Can rare events explain the equity premium puzzle? Review of Financial Studies. 2012; 25 (10): 3037–76. https://doi.org/10.1093/rfs/hhs078.
- View Article
- Google Scholar
6. Barro RJ, Ursúa JF. Rare macroeconomic disasters. Annual Review of Economics. 2012; 4(1): 83–109. https://doi.org/10.1146/annurev-economics-080511-110932.
- View Article
- Google Scholar
7. Gabaix X. Variable rare disasters: an exactly solved framework for ten puzzles in macro-finance. The Quarterly Journal of Economics. 2012; 127(2): 645–700. https://doi.org/10.1093/qje/qjs001.
- View Article
- Google Scholar
8. Wachter JA. Can the time-varying risk of rare disasters explain aggregate stock market volatility? The Journal of Finance. 2013; 68(3): 987–1035. https://doi.org/10.3386/w14386.
- View Article
- Google Scholar
9. Nakamura E, Steinsson J, Barro RJ, Ursúa JF. Crises and recoveries in an empirical model of consumption disasters. American Economic Journal: Macroeconomics. 2013; 5(3): 35–74. https://doi.org/10.1257/mac.5.3.35.
- View Article
- Google Scholar
10. Farhi E, Gabaix X. Rare disasters and exchange rates. The Quarterly Journal of Economics. 2016; 131(1): 1–52. https://doi.org/10.1093/qje/qjv040.
- View Article
- Google Scholar
11. Martin IW. Consumption-based asset pricing with higher cumulants. The Review of Economic Studies. 2013; 80(2): 745–773. https://doi.org/10.1093/restud/rds029.
- View Article
- Google Scholar
12. Marfè R, Pénasse J. Measuring macroeconomic tail risk. Paris December 2017 Finance Meeting EUROFIDAI-AFFI. 2021.
- View Article
- Google Scholar
13. Irarrazabal A, Parra-Alvarez JC. Time-varying disaster risk models: An empirical assessment of the Rietz-Barro hypothesis. 2015. Working Paper.
- View Article
- Google Scholar
14. Gourio F. Disaster risk and business cycles. American Economic Review. 2012; 102(6): 2734–66. https://doi.org/10.2139/ssrn.1573097.
- View Article
- Google Scholar
15. Gourio F. Credit risk and disaster risk. American Economic Journal: Macroeconomics. 2013; 5(3): 1–34. https://doi.org/10.2139/ssrn.2182203.
- View Article
- Google Scholar
16. Lu YK, Siemer M. Asset pricing with learning about disaster risk. 2011. Boston University Working paper.
- View Article
- Google Scholar
17. Veronesi P. The peso problem hypothesis and stock market returns. Journal of Economic Dynamics and Control. 2004; 28(4): 707–725. https://doi.org/10.1016/S0165-1889(03)00041-1.
- View Article
- Google Scholar
18. Vissing-Jørgensen A. Limited asset market participation and the elasticity of intertemporal substitution. Journal of Political Economy. 2002; 110(4): 825–853. https://doi.org/10.1086/340782.
- View Article
- Google Scholar
19. Zhang J, Kong D, Liu H, Wu J. Asset pricing with time varying pessimism and rare disasters. International Review of Economics & Finance. 2019; 60: 165–175. https://doi.org/10.1016/j.iref.2018.11.005.
- View Article
- Google Scholar
20. Wachter JA, Zhu Y. Learning with rare disasters. 2019. Working Paper. https://doi.org/10.2139/ssrn.3407397.
- View Article
- Google Scholar
21. Liu B, Niu Y, Yang J. Time‐varying risk of rare disasters, investment, and asset pricing. Financial Review. 2020; 55(3): 503–524. https://doi.org/10.1111/fire.12226.
- View Article
- Google Scholar
22. Welch I. The (time-varying) importance of disaster risk. Financial Analysts Journal. 2016; 72(5): 14–30. https://doi.org/10.2469/faj.v72.n5.3.
- View Article
- Google Scholar
23. Du D, Elkamhi R. Learning About Rare Economic Disasters and Asset Prices. 2012. Working paper.
- View Article
- Google Scholar
24. Gillman M, Kejak M, Pakoš M. Learning about rare disasters: implications for consumption and asset prices. Review of Finance. 2015; 19(3): 1053–1104. https://doi.org/10.1093/rof/rfu016.
- View Article
- Google Scholar
25. Cox JC, Ingersoll Jr JE, Ross SA. A theory of the term structure of interest rates. Econometrica. 1985; 53: 385–408. https://doi.org/10.1142/9789812701022_0005.
- View Article
- Google Scholar
26. Gabaix X. Linearity-generating processes: A modelling tool yielding closed forms for asset prices. 2007. Working Paper. New York University.
27. Bansal R, Ochoa M, Kiku D. Climate change and growth risk. 2017. NBER Working Paper. No. 23009.
- View Article
- Google Scholar
28. Barro RJ, Jin T. Rare events and long-run risk. Review of Economic Dynamics. 2021; 39: 1–25. https://doi.org/10.1016/j.red.2020.08.002.
- View Article
- Google Scholar
29. Bansal R, Yaron A. Risks for the long run: a potential resolution of asset pricing puzzles. The Journal of Finance. 2004; 59(4): 1481–1509. https://doi.org/10.1111/j.1540-6261.2004.00670.x.
- View Article
- Google Scholar
30. Barro RJ, Jin T. On the size distribution of macroeconomic disasters. Econometrica. 2011; 79(5): 1567–1589. https://doi.org/10.3982/ECTA8827.
- View Article
- Google Scholar
31. Chen H, Pakos M. Rational overreaction and underreaction in fixed income and equity markets: The role of time-varying timing premium. 2007. Working Paper.
- View Article
- Google Scholar
32. Weil P. Nonexpected utility in macroeconomics. The Quarterly Journal of Economics. 1990; 105(1): 29–42. https://doi.org/10.2307/2937817.
- View Article
- Google Scholar
33. Epstein LG, Zin SE. Substitution, risk aversion, and the temporal behavior of consumption and asset returns: An empirical analysis. Journal of Political Economy. 1991; 99(2): 263–286. https://doi.org/10.1086/261750.
- View Article
- Google Scholar
34. Hall RE. Intertemporal substitution in consumption. Journal of Political Economy. 1988; 96(2): 339–357. https://doi.org/10.1086/261539.
- View Article
- Google Scholar
35. Campbell JY. Consumption-based asset pricing. Handbook of the Economics of Finance. 2003; 1: 803–887. https://doi.org/10.2139/ssrn.343784.
- View Article
- Google Scholar
36. Bansal R, Kiku D, Yaron A. Risk for the long run: Estimation and inference. 2007. Working Paper. Duke University and the University of Pennsylvania.
37. Guvenen F. A parsimonious macroeconomic model for asset pricing. Econometrica. 2009; 77(6): 1711–1750. https://doi.org/10.3982/ECTA6658.
- View Article
- Google Scholar
38. Colacito R, Croce MM. International asset pricing with recursive preferences. The Journal of Finance. 2013; 68(6): 2651–2686. https://doi.org/10.1111/jofi.12088.
- View Article
- Google Scholar
39. Gourio F, Siemer M, Verdelhan A. International risk cycles. Journal of International Economics. 2013; 89(2): 471–484. https://doi.org/10.1016/j.jinteco.2011.10.001.
- View Article
- Google Scholar
40. Colacito R, Croce MM. Risks for the long run and the real exchange rate. Journal of Political Economy. 2011; 119(1): 153–181. https://doi.org/10.1086/659238.
- View Article
- Google Scholar
41. Abel AB. Risk premia and term premia in general equilibrium. Journal of Monetary Economics. 1999; 43(1): 3–33. https://doi.org/10.3386/w6683.
- View Article
- Google Scholar

[ref1] 1. Barro RJ. Rare disasters and asset markets in the twentieth century. The Quarterly Journal of Economics. 2006; 121(3): 823–866. https://doi.org/10.1162/qjec.121.3.823.
View Article
Google Scholar

[2] View Article

[3] Google Scholar

[ref2] 2. Barro RJ, Ursúa JF. Macroeconomic crises since 1870. National Bureau of Economic Research. 2008. https://doi.org/10.3386/w13940.
View Article
Google Scholar

[5] View Article

[6] Google Scholar

[ref3] 3. Gourio F. Disasters and recoveries. American Economic Review. 2008; 98(2): 68–73. https://doi.org/10.1257/aer.98.2.68.
View Article
Google Scholar

[8] View Article

[9] Google Scholar

[ref4] 4. Barro RJ. Rare disasters, asset prices, and welfare costs. American Economic Review. 2009; 99 (1): 243–64. https://doi.org/10.1257/aer.99.1.243.
View Article
Google Scholar

[11] View Article

[12] Google Scholar

[ref5] 5. Julliard C, Anisha G. Can rare events explain the equity premium puzzle? Review of Financial Studies. 2012; 25 (10): 3037–76. https://doi.org/10.1093/rfs/hhs078.
View Article
Google Scholar

[14] View Article

[15] Google Scholar

[ref6] 6. Barro RJ, Ursúa JF. Rare macroeconomic disasters. Annual Review of Economics. 2012; 4(1): 83–109. https://doi.org/10.1146/annurev-economics-080511-110932.
View Article
Google Scholar

[17] View Article

[18] Google Scholar

[ref7] 7. Gabaix X. Variable rare disasters: an exactly solved framework for ten puzzles in macro-finance. The Quarterly Journal of Economics. 2012; 127(2): 645–700. https://doi.org/10.1093/qje/qjs001.
View Article
Google Scholar

[20] View Article

[21] Google Scholar

[ref8] 8. Wachter JA. Can the time-varying risk of rare disasters explain aggregate stock market volatility? The Journal of Finance. 2013; 68(3): 987–1035. https://doi.org/10.3386/w14386.
View Article
Google Scholar

[23] View Article

[24] Google Scholar

[ref9] 9. Nakamura E, Steinsson J, Barro RJ, Ursúa JF. Crises and recoveries in an empirical model of consumption disasters. American Economic Journal: Macroeconomics. 2013; 5(3): 35–74. https://doi.org/10.1257/mac.5.3.35.
View Article
Google Scholar

[26] View Article

[27] Google Scholar

[ref10] 10. Farhi E, Gabaix X. Rare disasters and exchange rates. The Quarterly Journal of Economics. 2016; 131(1): 1–52. https://doi.org/10.1093/qje/qjv040.
View Article
Google Scholar

[29] View Article

[30] Google Scholar

[ref11] 11. Martin IW. Consumption-based asset pricing with higher cumulants. The Review of Economic Studies. 2013; 80(2): 745–773. https://doi.org/10.1093/restud/rds029.
View Article
Google Scholar

[32] View Article

[33] Google Scholar

[ref12] 12. Marfè R, Pénasse J. Measuring macroeconomic tail risk. Paris December 2017 Finance Meeting EUROFIDAI-AFFI. 2021.
View Article
Google Scholar

[35] View Article

[36] Google Scholar

[ref13] 13. Irarrazabal A, Parra-Alvarez JC. Time-varying disaster risk models: An empirical assessment of the Rietz-Barro hypothesis. 2015. Working Paper.
View Article
Google Scholar

[38] View Article

[39] Google Scholar

[ref14] 14. Gourio F. Disaster risk and business cycles. American Economic Review. 2012; 102(6): 2734–66. https://doi.org/10.2139/ssrn.1573097.
View Article
Google Scholar

[41] View Article

[42] Google Scholar

[ref15] 15. Gourio F. Credit risk and disaster risk. American Economic Journal: Macroeconomics. 2013; 5(3): 1–34. https://doi.org/10.2139/ssrn.2182203.
View Article
Google Scholar

[44] View Article

[45] Google Scholar

[ref16] 16. Lu YK, Siemer M. Asset pricing with learning about disaster risk. 2011. Boston University Working paper.
View Article
Google Scholar

[47] View Article

[48] Google Scholar

[ref17] 17. Veronesi P. The peso problem hypothesis and stock market returns. Journal of Economic Dynamics and Control. 2004; 28(4): 707–725. https://doi.org/10.1016/S0165-1889(03)00041-1.
View Article
Google Scholar

[50] View Article

[51] Google Scholar

[ref18] 18. Vissing-Jørgensen A. Limited asset market participation and the elasticity of intertemporal substitution. Journal of Political Economy. 2002; 110(4): 825–853. https://doi.org/10.1086/340782.
View Article
Google Scholar

[53] View Article

[54] Google Scholar

[ref19] 19. Zhang J, Kong D, Liu H, Wu J. Asset pricing with time varying pessimism and rare disasters. International Review of Economics & Finance. 2019; 60: 165–175. https://doi.org/10.1016/j.iref.2018.11.005.
View Article
Google Scholar

[56] View Article

[57] Google Scholar

[ref20] 20. Wachter JA, Zhu Y. Learning with rare disasters. 2019. Working Paper. https://doi.org/10.2139/ssrn.3407397.
View Article
Google Scholar

[59] View Article

[60] Google Scholar

[ref21] 21. Liu B, Niu Y, Yang J. Time‐varying risk of rare disasters, investment, and asset pricing. Financial Review. 2020; 55(3): 503–524. https://doi.org/10.1111/fire.12226.
View Article
Google Scholar

[62] View Article

[63] Google Scholar

[ref22] 22. Welch I. The (time-varying) importance of disaster risk. Financial Analysts Journal. 2016; 72(5): 14–30. https://doi.org/10.2469/faj.v72.n5.3.
View Article
Google Scholar

[65] View Article

[66] Google Scholar

[ref23] 23. Du D, Elkamhi R. Learning About Rare Economic Disasters and Asset Prices. 2012. Working paper.
View Article
Google Scholar

[68] View Article

[69] Google Scholar

[ref24] 24. Gillman M, Kejak M, Pakoš M. Learning about rare disasters: implications for consumption and asset prices. Review of Finance. 2015; 19(3): 1053–1104. https://doi.org/10.1093/rof/rfu016.
View Article
Google Scholar

[71] View Article

[72] Google Scholar

[ref25] 25. Cox JC, Ingersoll Jr JE, Ross SA. A theory of the term structure of interest rates. Econometrica. 1985; 53: 385–408. https://doi.org/10.1142/9789812701022_0005.
View Article
Google Scholar

[74] View Article

[75] Google Scholar

[ref26] 26. Gabaix X. Linearity-generating processes: A modelling tool yielding closed forms for asset prices. 2007. Working Paper. New York University.

[ref27] 27. Bansal R, Ochoa M, Kiku D. Climate change and growth risk. 2017. NBER Working Paper. No. 23009.
View Article
Google Scholar

[78] View Article

[79] Google Scholar

[ref28] 28. Barro RJ, Jin T. Rare events and long-run risk. Review of Economic Dynamics. 2021; 39: 1–25. https://doi.org/10.1016/j.red.2020.08.002.
View Article
Google Scholar

[81] View Article

[82] Google Scholar

[ref29] 29. Bansal R, Yaron A. Risks for the long run: a potential resolution of asset pricing puzzles. The Journal of Finance. 2004; 59(4): 1481–1509. https://doi.org/10.1111/j.1540-6261.2004.00670.x.
View Article
Google Scholar

[84] View Article

[85] Google Scholar

[ref30] 30. Barro RJ, Jin T. On the size distribution of macroeconomic disasters. Econometrica. 2011; 79(5): 1567–1589. https://doi.org/10.3982/ECTA8827.
View Article
Google Scholar

[87] View Article

[88] Google Scholar

[ref31] 31. Chen H, Pakos M. Rational overreaction and underreaction in fixed income and equity markets: The role of time-varying timing premium. 2007. Working Paper.
View Article
Google Scholar

[90] View Article

[91] Google Scholar

[ref32] 32. Weil P. Nonexpected utility in macroeconomics. The Quarterly Journal of Economics. 1990; 105(1): 29–42. https://doi.org/10.2307/2937817.
View Article
Google Scholar

[93] View Article

[94] Google Scholar

[ref33] 33. Epstein LG, Zin SE. Substitution, risk aversion, and the temporal behavior of consumption and asset returns: An empirical analysis. Journal of Political Economy. 1991; 99(2): 263–286. https://doi.org/10.1086/261750.
View Article
Google Scholar

[96] View Article

[97] Google Scholar

[ref34] 34. Hall RE. Intertemporal substitution in consumption. Journal of Political Economy. 1988; 96(2): 339–357. https://doi.org/10.1086/261539.
View Article
Google Scholar

[99] View Article

[100] Google Scholar

[ref35] 35. Campbell JY. Consumption-based asset pricing. Handbook of the Economics of Finance. 2003; 1: 803–887. https://doi.org/10.2139/ssrn.343784.
View Article
Google Scholar

[102] View Article

[103] Google Scholar

[ref36] 36. Bansal R, Kiku D, Yaron A. Risk for the long run: Estimation and inference. 2007. Working Paper. Duke University and the University of Pennsylvania.

[ref37] 37. Guvenen F. A parsimonious macroeconomic model for asset pricing. Econometrica. 2009; 77(6): 1711–1750. https://doi.org/10.3982/ECTA6658.
View Article
Google Scholar

[106] View Article

[107] Google Scholar

[ref38] 38. Colacito R, Croce MM. International asset pricing with recursive preferences. The Journal of Finance. 2013; 68(6): 2651–2686. https://doi.org/10.1111/jofi.12088.
View Article
Google Scholar

[109] View Article

[110] Google Scholar

[ref39] 39. Gourio F, Siemer M, Verdelhan A. International risk cycles. Journal of International Economics. 2013; 89(2): 471–484. https://doi.org/10.1016/j.jinteco.2011.10.001.
View Article
Google Scholar

[112] View Article

[113] Google Scholar

[ref40] 40. Colacito R, Croce MM. Risks for the long run and the real exchange rate. Journal of Political Economy. 2011; 119(1): 153–181. https://doi.org/10.1086/659238.
View Article
Google Scholar

[115] View Article

[116] Google Scholar

[ref41] 41. Abel AB. Risk premia and term premia in general equilibrium. Journal of Monetary Economics. 1999; 43(1): 3–33. https://doi.org/10.3386/w6683.
View Article
Google Scholar

[118] View Article

[119] Google Scholar

Figures

Abstract

Introduction

Rare economic disasters

The model

Proposition 1

Asset return

Stochastic discount factor

Assumption 1.

Proposition 2.

Proposition 3.

Proposition 4.

The risk-free rate

The equity return

Proposition 5.

Calibration and simulation

Calibration

Simulation results

Implied disaster probability

Predictability

Conclusions

Supporting information

S1 Data.

S1 Appendix.

References