Semi-coskewnesses and the cross-section of excepted stock returns: Evidence from China
Abstract
We propose an alternative nonlinear semi-risk measure, by decomposing the traditional coskewness into four components associated with the signed excess market and asset returns, that captures the asymmetries in nonlinear markets. We find that the two semi-coskewnesses attributable to (positive) negative excess market returns predict significantly (lower) higher future returns based on high-frequency data from China’s A-share market. After conducting a wide range of implementations, the risk premium for negative coskewness stands out as the most significant, followed by the premium for mixed negative coskewness. In contrast, the results for positive and mixed positive coskewnesses are not always significantly negative. More importantly from an economically meaningful perspective, for a downside risk premium of 25.40% per annum, a 2-standard-deviation increase in negative semi-coskewness gives rise to an increase of approximately 13.71% in annual expected return.
https://doi.org/10.5061/dryad.80gb5mkx7
Our empirical data consists of high-frequency intraday data on the opening and closing prices of all stocks (shares issued to domestic investors) from the China A-share stock market, along with weekly data on equity-related indicators.
Description of the data and file structure
The sample period selected in this paper spans from April 8, 2005 to April 22, 2022, with daily trading hours from 9:30 to 11:30 and 13:00 to 15:00 in GMT+8. Since too high data frequency results to larger market microscopic noise, and too low data frequency makes it difficult to capture high-frequency signals, Liu, Patton and Sheppard (2015) provide evidence that the estimations of realized measures based on a 5-minute data frequency perform well in terms of prediction accuracy, and a number of studies (Patton and Sheppard, 2015; Amaya, Christoffersen, Jacobs, and Vasquez, 2015; Bollerslev, 2020) adopt this frequency. Therefore, we choose a five-minute data frequency as well. We calculate daily realized semi-coskewnesses using 5-minute intraday returns and aggregate them to obtain weekly frequency. We also extract market capitalization, turnover rate, and book-to-market ratio for each stock from the CRSP database and the RESSET Financial Research database respectively. Lastly, we use daily returns to compute weekly returns, realized (co)moments, realized semi-risk factors, and lagged returns as well as maximum/minimum returns over the previous month.
Each CSV file contains weekly measures for the semi-coskewness factors and firm characteristics of each stock over a year, e.g., a CSV file named 2005 contains weekly measures for the semi-coskewness factors and firm characteristics of each stock during 2005. In addition, each file's column headers are described as follows:
- T: each week
- ID: stock code
- R: the stock return over week t.
- RVOL: the weekly realized volatility
- RVOL+ and RVOL-: the weekly realized good and bad volatilities
- RSK: the weekly realized skewness
- RSK+ and RSK-: the weekly realized upside and downside skewensses
- RKT: the weekly realized kurtosis
- CSK: the weekly realized coskewness
- CSK+ and CSK-: the weekly realized upside and downside coskewensses
- CSKP, CSKM+, CSKN and CSKM-: the four different weekly realized semi-coskewnesses (positive, mixed positive, negative, and mixed negative coskewnesses)
- CKT: the weekly realized kurtosis
- R1: the stock return over week t+1
- TO: turnover rate
- BM: book-to-market ratio
- ME: the logarithm of the market capitalization of the firm
- REV: the previous week’s return
- MAX: the maximum daily returns over the previous month
- MIN: the minimum daily returns over the previous month
- BETA: the weekly realized beta
- BETA+ and BETA-: the weekly realized upside and downside betas
- BETAP, BETAM+, BETAN and BETAM-: the four different realized semibetas
- IVOL: idiosyncratic volatility.
- MOM: momentum
Sharing/Access information
Data was derived from the following sources:
- CRSP database
- RESSET Financial Research database
Code/Software
The Code folder contains the source code for all empirical study presented in the paper. The files containing the string ".m" are implemented in MATLAB, while the file containing the string ".R" is implemented in R.
We calculate daily realized semi-coskewnesses using 5-minute intraday returns and aggregate them to obtain weekly frequency based on all the listed stocks of China’s A-share stock market. We also extract market capitalization, turnover rate, and book-to-market ratio for each stock from the CRSP database and the RESSET Financial Research database respectively. Lastly, we use daily returns to compute weekly returns, realized (co)moments, realized semi-risk factors, and lagged returns as well as maximum/minimum returns over the previous month.