External shocks to the stock market make investors highly aware of the risk of asset investment and lead them to expect correspondingly higher returns. As a result, risk in the market increases, causing asset prices to fall. Applying a Granger causality test to data from the COVID-19 pandemic period, this study tested whether the COVID-19 fear index is useful in predicting asset prices in the stock and cryptocurrency markets and assessed the COVID-19 fear index's Long Short-Term Memory model-based asset price prediction performance. The Long Short-Term Memory model was developed to deal with the vanishing gradient problem that can emerge when training traditional Recurrent Neural Networks using the cell states. Using data spanning the period from early 2020 when COVID-19 began to April 2022, this study's empirical analysis produced the following results. First, it showed significant negative correlations between the fear index and both stock and cryptocurrency prices. During the COVID-19 pandemic crisis, cryptocurrencies played the role of speculative rather than safe assets. Second, both stock and cryptocurrency prices showed significant Granger causality. Third, the impulse response function indicated that both stocks and cryptocurrencies overreacted to the shock of the COVID-19 pandemic. The degree of overreaction was stronger in the stock market. Fourth, the price prediction performance of the Long Short-Term Memory time series model using price data and the COVID-19 fear index as input variables was excellent. In particular, its forecast performance for medium-and small-sized assets was higher than for large-sized assets such as Bitcoin and largesized KOSPI. While outbreaks of infectious diseases such as COVID-19 directly impact the economy and the stock market, this study's results indicate that COVID-19 had less of an impact on the cryptocurrency market, which is also affected by non-economic factors.