Financial markets are a wellspring of quantifiable data, but the data generation process remains a deep mystery. Using daily stock market returns naively can lead to an underestimation of the risk of extended sell-offs similar to the February-March 2020 episode when the S&P 500 Index tumbled 34% over 23 trading days. An empirical analysis comparing the inferences gained from historical 1928-2019 daily, monthly & quarterly returns data of the S&P 500 Index.
The analysis reveals that the volatility clustering and aggregational normality phenomena suggest that historical daily returns statistics underestimate the risk of extended volatility spikes in the S&P 500 Index, and significant drawdowns of 30% or more over a month. Monthly 1928โ2019 historical data would have provided higher probabilities of a scenario akin to the February-March 2020 crash, than using historical daily returns data. More recent daily data from the 1990โ2019 period would have failed to foresee such an eventuality altogether.
This is the Python code used for the article published in: https://towardsdatascience.com/simulating-stock-market-crash-risk-85aae5abb197