Financial markets aren’t static. In good times, riskier assets are expected to perform well, compensating investors for systematic risk. But during bad times, like bear markets, these same assets often underperform. Understanding this dynamic – how the relationship between risk and return changes with the market’s mood – is crucial for investors.
Contents
Traditional financial models often struggle to fully capture these shifts. Many define bad states using pre-set thresholds or assume the market state is already known when prices are determined. This can lead to inaccurate predictions, especially during turbulent periods.
A recent study by Massacci, Sarno, and Trapani introduces a novel conditional asset pricing model that addresses these limitations. This model uses bear market risk – the forward-looking probability of a future bear market – as the key factor determining the market state. By doing so, it offers a more nuanced understanding of asset pricing across different economic conditions.
Key takeaways from this new model include:
- Improved Accuracy: It predicts asset returns more accurately than traditional unconditional models, especially during distinct good and bad market states.
- State-Dependent Risk Premiums: It shows that the compensation investors require for taking on certain risks (risk premia) can vary significantly depending on whether the market is in a good or bad state.
- Changing Diversification Benefits: The structure of risk factors appears to change across states, suggesting that diversification benefits may diminish precisely when investors need them most – in bad times.
The Challenge of Changing Market States
Asset pricing models explain how expected returns relate to risk. However, the exposure of assets to common risks (like market risk, interest rate risk, etc.) and the compensation investors demand for these risks (risk premia) can change over time and across different market conditions. Research suggests stocks tend to move together more strongly in bad times, reducing the benefits of diversification.
Existing “downside risk” models acknowledge the existence of bad states, often triggered when the overall market falls below a specific level (e.g., a -3% return). While valuable, these models typically:
- Assume the market state is known when investors make decisions.
- Use an assumed, rather than estimated, threshold for defining the bad state.
These assumptions can limit their ability to capture the real-time, conditional nature of market dynamics.
A Conditional Approach Using Bear Market Risk
The model developed by Massacci, Sarno, and Trapani (2025) is conditional, meaning asset pricing is based on the information available to investors at the time prices are set. Its core innovation is using bear market risk as the conditioning variable. This isn’t just looking at past returns but the expected probability of a future bear market, derived from market indicators like prices of S&P 500 options. The return on an “Arrow-Debreu Bear portfolio” – which pays out if the market enters a bear state – serves as a forward-looking measure of this risk.
Unlike previous models, this framework:
- Estimates the Threshold: It doesn’t assume the threshold level that triggers a state change is known beforehand; instead, it estimates this value from the data.
- Uses Latent Factors: It employs a “latent factor” approach to identify the systematic risk factors present in each state. This avoids the problem of potentially omitting important observable factors and allows for estimating risk premia for any candidate factor, observable or not.
This sophisticated methodology provides a powerful tool for understanding the complex, state-dependent nature of asset returns and risk pricing.
Empirical Evidence: Better Pricing in US Equities
Applying this conditional model to US equity portfolio returns reveals significant insights. The expectation of a future bear market, driven by the Arrow-Debreu Bear portfolio return, acts as the mechanism determining whether the market is in a “good” or “bad” state for pricing purposes in the next period.
Empirically, the difference between the new conditional model and traditional unconditional models is striking. An unconditional model, which ignores the changing market state, performs poorly, systematically overestimating returns in bad states and underestimating them in good states. This results in large pricing errors, even if the average error across all periods looks small.
Comparison of realized equity portfolio returns: Unconditional average versus conditional returns in good and bad market states.
Figure 1 clearly shows how unconditional models misprice assets by averaging across different market regimes. Returns in good states are much higher than the unconditional average, while returns in bad states are much lower.
In contrast, the conditional latent factor model closely tracks realized returns in both good and bad states, showing no systematic pricing errors.
Comparison of predicted and realized equity returns demonstrating the accuracy of the conditional asset pricing model.
As illustrated in Figure 2, the conditional model’s predicted returns align well with realized returns, both within the sample period and when tested on out-of-sample data. The average pricing errors for the conditional model are significantly smaller – about one-tenth the size of those from the unconditional model.
Furthermore, the study found that a larger number of common factors are needed to explain returns in good states compared to bad states. This supports the idea that assets move more independently in bull markets, offering diversification benefits, while they become more correlated and driven by fewer factors during bear markets.
Risk Premia Shift with the Market Tide
The model also allows researchers to estimate the risk premium for any factor (observable or not) in both good and bad states. Consider the VIX index, often used as a measure of market volatility risk. The change in VIX (∆VIX) is known to be a priced risk factor. Since volatility typically spikes in bad times, assets that perform poorly when VIX rises are riskier and should demand a positive risk premium (or equivalently, ∆VIX should carry a negative risk premium).
Using their conditional model, the authors estimated the risk premium for ∆VIX:
- In the good state: The risk premium was -0.92% per annum, consistent with previous unconditional estimates.
- In the bad state: The risk premium was statistically zero.
This finding is significant. It quantifies how the expected compensation for taking on volatility risk essentially disappears during bear markets, confirming that risk-return relationships are not constant but depend heavily on the prevailing market regime.
Conclusion: A Powerful Tool for State-Dependent Analysis
The research presents a robust new methodology for modeling and pricing assets by explicitly accounting for the dynamic nature of market states, driven by forward-looking bear market risk. By estimating the threshold that determines state changes and employing a latent factor approach, the model overcomes key limitations of previous methods.
The empirical results for US equities demonstrate its superior performance in capturing the cross-section of returns across good and bad market states. It also provides valuable insights into how factor structures and risk premia change with market conditions.
This general framework has potential applications beyond basic asset pricing, including advanced risk management techniques like analyzing CoVaR (conditional value at risk) or understanding interconnected risks during market crises. For investors and analysts, this model underscores the importance of considering market states and the conditional nature of risk and return when evaluating investment opportunities.
For more research on market risk and asset pricing, explore related articles on [insert link to related content category or page, e.g., financial modeling or market risk analysis].
