What have we learned from 20 million historical US stock data?

Abstract

This study aims to statistically characterize the US stock market from January 3, 1995 to June 11, 2021. The literature is primarily based on either limited data or certain major indexes such as the Standard & Poor’s 500, Nasdaq or Dow Jones. Our analyses use around 20 million end-of-day data points for 9701 stocks traded in the United States. Specifically, we investigate calendar and correlation effects on stock returns. Our aim is to provide practical analyses of the US stock market as well as investment strategies for traders and investors. This paper answers a series of questions, such as: which months or days of the week have led to the worst/best returns and what their monthly/daily volatility looks like; whether the stock market’s behavior has changed over the years due to the popularity of exchange-traded funds and, if so, what the implications are; statistically the best day on which to buy or sell if someone wishes to hold stock for four days; whether it is true that day trading is not advantageous and, if so, why; and finally, what the historical returns for the daily, weekly and monthly trades were.