Package JumpTest usage tips

Overview

This vignette provides an introduction to nonparametric tests to financial jump detection. R package JumpTest provided both simulation methods and test methods with multiple tests correction. The package can be loaded with command:

library(JumpTest)

## Loading required package: MASS

Workflow: Meta Analysis

This package provided analysis on high frequency stock markets. You can simply simulate the data using the simulation codes we provided. For example you can simulate a stoachasitc volatility with one factor and jump (SV1FJ) model (Dumitru and Urga 2012) for 1200 days using the following code.

M <- 390*60
m <- 1200
pric <- SV1FJ(M,m,3,.2,.03,.5,0,.125,-.1,-.62)
lp <- c(3,pric[,1])
lr <- diff(lp[seq(1,M*m+1,60)])
logreturn <- matrix(lr,ncol=m)

Here we simulated our data by seconds, while we grabbed the minute level data and saved the log returns per day by columns. Then we can used the following code to obtain the p-values for three methods (BNS (Barndorff-Nielsen, Shephard, and Winkel 2006), Amed, and Amin (Andersen, Dobrev, and Schaumburg 2012)).

pmat <- pcombine(logreturn,c('BNS','Amed','Amin'))

Then we can obtain the p-value matrix for the three methods. We can do further p-values pooling and FDR control via BH (Benjamini and Hochberg 1995, @yen2013testing) procedure. We provided six pooling methods, four for independent (Won et al. 2009, @chang2013meta) and two for dependent (Chang et al. 2013) strucure. We suggested Stouffer’s dependent (SD) structure.

res <- ppool(pmat)
pvalue <- res@pvalue
pvalue_adjusted <- res@adjp

Conclusion and Ongoing Work

R package JumpTest provides effective tools to detect jump in high frequency stock market data. It can efficiently identify intervals that contain jumps. In the future we may develop methods that can locate the exact location of jumps.