Datasets for testing the performances of jump diffusion models

This article contains datasets related to the research article titled a novel jump diffusion model based on SGT distribution and its applications (”A novel jump diffusion model based on SGT distribution and its applications” (W.J. Xu, G.F. Liu, H.Y. Li, 2016) [1]). The datasets contain continuous composite daily percentage return values which are computed from the daily closing prices. Firstly, we describe statistical properties of the datasets. Then, the datasets are split into two samples, the in-sample data and out-of-sample data. The datasets can be used as benchmarks for testing the performances of jump diffusion models.


a b s t r a c t
This article contains datasets related to the research article titled a novel jump diffusion model based on SGT distribution and its applications ("A novel jump diffusion model based on SGT distribution and its applications" (W.J. Xu, G.F. Liu, H.Y. Li, 2016) [1]). The datasets contain continuous composite daily percentage return values which are computed from the daily closing prices. Firstly, we describe statistical properties of the datasets. Then, the datasets are split into two samples, the in-sample data and out-ofsample data. The datasets can be used as benchmarks for testing the performances of jump diffusion models.
& 2017 Published by Elsevier Inc. This is an open access article under the CC BY license (http://creativecommons.org/licenses/by/4.0/).

Subject area
Economics More specific subject area

Financial Engineering
Type of data

Experimental factors
In order to the empirical research, the dataset is split into two samples, the insample data and out-of-sample data.

Experimental features
The data is the daily percentage return values of four representative composite indices and is public data in financial market.

Guangzhou, China
Data accessibility Data is within this article (http://www.wind.com.cn/Default.aspx)

Value of the data
The data is convenient to execute the statistical analysis and empirical application in this paper. The data can be used to test the existence of jumps in four representative composite indices and estimate the relevant model parameters.
The data can be used to assess the asset return distribution describing performance of relevant models.
The data can be used to explore the volatility forecast performance of relevant models based on insample data and out-of-sample data respectively.

Data
The raw data contains the daily closing price of four representative composite indices (the Nikkei 225 Index (NIKKEI225), the Dow Jones Industrial Average Index (DJIA), Hang Seng Composite Index (HSI), and the Shanghai Composite Index (SCI)). The time period is from January 3, 1995 to March 25, 2016.
In order to explore the performance of jump diffusion models, the daily closing price is converted into daily percentage return value.

Experimental design, materials and methods
The datasets, daily closing price time series of asset S t ðt ¼ 1; 2; …; NÞ, are obtained from the Wind Finance Database (http://www.wind.com.cn/Default.aspx) in China. In order to explore the asset return distribution describing performance of jump diffusion models, the datasets are converted into daily percentage return values y t by using the following equation: where lnðS t Þ is the natural logarithm of the closing price S t at t. All the datasets are listed in Table 1. The daily closing prices and daily percentage return values are shown in Supplementary materials (data.xlsx).
Finally, on the performance of volatility forecasts, several GARCH family models with some compound return distributions are presented. The datasets are split into two samples, the in-sample data and out-of-sample data (see Fig. 1). In order to compare the performance of volatility forecasts of relevant models, we use the rolling-window approach (One step forward). The initial time period of in-of-sample data is from January 3, 1995 to 26 April, 2013. For each data series, these relevant models are first estimated using the in-of-sample data (before the time t), and a volatility value is obtained as a forecast volatility at the next time tþ1 (see Fig. 1). Subsequently, the estimation period was rolled forward by adding one new day. By repeating this procedure, the out-of-sample volatility forecasts were calculated for the rest days.

Transparency document. Supporting information
Transparency data associated with this article can be found in the online version at http://dx.doi. org/10.1016/j.dib.2016.11.014.

Appendix A. Supporting information
Supplementary data associated with this article can be found in the online version at http://dx.doi. org/10.1016/j.dib.2016.11.014.