What drives stocks during the Corona Crash? News Attention vs Rational Expectation

Reading Time: 4 minutes

As already outlined in our previous article (which you can read here), the uncertainty resulting from the COVID-19 pandemic led to a dramatic downturn in global financial markets in the first quarter of the year. For instance, the S&P500 dropped by 33% from its all-time high over the period from January to March 2020.

But what was the reason for this dramatic drop? Were investors really focusing on rational expectations about the pandemic’s development or were they rather driven by the news hype when making their investment decisions?

These questions are at the heart of our recent study called “What Drives Stocks during the Corona-Crash? News Attention vs. Rational Expectation” (for more information, see here). In this study, we examine a sample of 64 national stock markets, covering 94% of the world’s GDP, and find the stock markets’ decline to be mainly associated with higher news attention and less with rational expectation.

But how do we proxy for rational expectations and news attention?

To proxy for news attention, we calculate an abnormal Google search volume index (ASVI) as presented in Da et al. (2011). Although the ASVI is generally used to measure retail investor attention, Plante (2019) also shows that the search volume strongly correlates with news attention as the amount of news the public is confronted with translates into a rise of related Google searches. Therefore, the ASVI also seems to be an appropriate proxy for general news attention. However, we adjust the measure proposed by Da et al. (2011) to account for our shorter time window. As a result, our news coverage variable NCi,t is calculated as the natural log of the search volume on trading day t-1 minus the natural log of the median search volume over the previous five trading days.

(1)   \begin{equation*}NC_i_,_t=log(SVI_t_-_1)-log(med(SVI_t_-_2,...,SVI_t_-_6))\end{equation*}

To proxy for rational expectations about the pandemic’s development, we turn to epidemiological models. Since the spread of infections is initially characterized by an exponential growth in time, we fit the number of infections per country to an exponential growth model using data from Johns Hopkins University. We then calculate daily exponential growth rates by fitting the exponential growth model; and use the change in growth rates per country between t and t+1 as a proxy for a rational investor’s expectation about the pandemic’s development in our regression models.

In later stages of the pandemic, and as countermeasures unfold, the exponential growth is typically weakened, and the infections start following a logistic function. Therefore, we also fit the epidemiological standard model – the Susceptible-Infectious-Recovered model (SIR). This model accounts for both the number of infected individuals and the number of susceptible and recovered individuals in a population. Based on the assumption of immunity of recovered individuals, the SIR model derives from a set of differential equations as the transmissions between the groups of individuals are formulated as derivatives (for an overview, see e.g. Ma, 2020). Same as with the exponential growth model, we fit the SIR model at each time step and use the change in growth rates as an independent variable for a rational investor’s expectation in an additional regression model.

But how do we come to our results?

To examine whether the decline in stock prices was driven by investors’ rational expectations about the pandemic’s development or rather by news attention, we consider the following straightforward regression model:

(2)   \begin{equation*}MKT_i_,_t=\rho MKT_i_,_t_-_1+\beta _1EXP_i_,_t+\beta _2NC_i_,_t+\epsilon _i_,_t\end{equation*}


where A is the index return for country i on trading day t. EXP_i_,_t  is our measure for an investor’s rational expectation based on the exponential growth model, while NC_i_,_t measures news attention for the keyword “corona”. In an additional regression model, we use SIR_i_,_t ,which is our measure for rational expectation based on the SIR Model, as an independent variable instead of EXP_i_,_t. Yet, in both regressions we control for all other market effects using the lagged log returns of the national stock market indices denoted by MKT_i_,_t_-_1. Because of the latter, however, the ordinary least squares (OLS) estimator is biased and inconsistent since the regressors are no longer exogenous. We therefore estimate the model using the generalized method of moments (GMM) estimator, which yields unbiased and consistent estimates (Arellano & Bond, 1991; Hansen, 1982). The results from these regressions are presented in the table below. To facilitate interpretation and to make comparisons possible, all variables are scaled to have a standard deviation of one and are weighted by their country’s GDP.

Dependent Variable: Market Return (MKT_i_,_t)

Model (1)

Model (2)

Lagged Market Return (MKT_i_,_t_-_1)





Expected Exponential Growth Rate (EXP_i_,_t)




Expected SIR Growth Rate (SIR_i_,_t)




News Attention (NC_i_,_t)











Trading Days



Estimation Method



Robust Standard Errors



Country Fixed Effects



Time Fixed Effects



*, **, *** denote significance at the 10%, 5%, 1% level

In Model 1, all regression coefficients are found to be negative and statistically significant. The larger coefficient on NC_i_,_t compared to EXP_i_,_t indicates that news attention is the dominant driver of the drop in stock prices over the entire observation period. To put this into perspective, a one standard deviation increase in news attention leads to a decrease of 0.279 standard deviations of market returns, while a one standard deviation increase in our rational expectation variable results in a decrease of 0.131 standard deviations of market returns. Furthermore, the coefficient on the one-day lagged market return is found to be larger in magnitude compared to our rational expectation variable. This implies that yesterday’s market development has a larger impact than the rational expectation for the next day.

In Model 2, we again find negative and statistically significant coefficients on NC_i_,_t and MKT_i_,_t_-_1, which are similar in size compared to Model 1. However, we do not find statistically significant coefficients on our rational expectation variable SIR_i_,_t. Hence, this provides further support indicating the stock markets’ decline to be mainly associated with higher news attention and less with rational expectation.

In conclusion, our first set of results presented in this blog post as well as our battery of robustness checks included in our paper provides convincing evidence indicating that the higher news attention was the dominant driver behind the downturn of the global financial markets at the beginning of the corona crisis.


Arellano, M., & Bond, S. (1991). Some tests of specification for panel data: Monte Carlo evidence and an application to employment equations. The review of economic studies, 58(2), 277-297.

Da, Z., Engelberg, J., & Gao, P. (2011). In search of attention. The Journal of Finance, 66(5), 1461-1499.

Hansen, L. P. (1982). Large sample properties of generalized method of moments estimators. Econometrica: Journal of the Econometric Society, 1029-1054.

Ma, J. (2020). Estimating epidemic exponential growth rate and basic reproduction number. Infectious Disease Modelling, 5, 129-141.Plante, M. (2019). OPEC in the News. Energy Economics, 80, 163-172.

Print Friendly, PDF & Email