報(bào)告承辦單位: 數(shù)學(xué)與統(tǒng)計(jì)學(xué)院
報(bào)告內(nèi)容: Testing for Series Correlation and ARCH Effect of High-Dimensional Time Series Data
報(bào)告人姓名: 凌仕卿
報(bào)告人所在單位: 香港科技大學(xué)
報(bào)告人職稱/職務(wù)及學(xué)術(shù)頭銜: 教授,博導(dǎo)
報(bào)告時(shí)間: 2018年10月17日周三下午3:30
報(bào)告地點(diǎn): 理科樓A419
報(bào)告人簡(jiǎn)介: 凌仕卿教授于1997年取得香港大學(xué)統(tǒng)計(jì)學(xué)博士學(xué)位,1997年至2000年西澳大學(xué)經(jīng)濟(jì)學(xué)系博士后,2000年至2006年香港科技大學(xué)數(shù)學(xué)系助理教授,2003年至2006年受聘于西澳大學(xué)經(jīng)濟(jì)學(xué)系和數(shù)學(xué)與統(tǒng)計(jì)系兼職副教授,2006年至2010年香港科技大學(xué)數(shù)學(xué)系副教授,2010年至今香港科技大學(xué)數(shù)學(xué)系教授。凌教授的主要研究方向?yàn)椋捍髽颖纠碚?、?jīng)驗(yàn)過(guò)程、非平穩(wěn)時(shí)間序列、非線性時(shí)間序列及計(jì)量經(jīng)濟(jì)學(xué)。現(xiàn)為《Journal of Time Series Analysis》聯(lián)合編輯《Statistics & Probability Letters》、《Bernoulli》、《Electronic Journal of Statistics》、《Journal of the Japan Statistical Association》國(guó)際期刊的副主編。2003年和2013年分別榮獲澳大利亞和新西蘭MSS委員會(huì)頒發(fā)的Early Career Research Excellence Prize和Biennial Medal, 2005年當(dāng)選為國(guó)際統(tǒng)計(jì)學(xué)會(huì)會(huì)員;2007年榮獲計(jì)量經(jīng)濟(jì)學(xué)期刊(Econometric Theory)頒發(fā)的Multa Scripsit Award 的獎(jiǎng)勵(lì),2013年當(dāng)選為澳大利亞和新西蘭MSS的Fellow。2015年當(dāng)選為ITTI的Inaugural Distinguished Fellow。
報(bào)告摘要:This paper proposes two Portmanteau tests for detecting serial correlation and ARCH effect in high-dimensional data. The dimension of data $p=p(n)$ may go to infinity when the sample size $n\to\infty$. We first show that the sample autocorrelation function of the $L_{1}-$norm of data is asymptotically normal and a norm-based Portmanteau test statistic is asymptotically $\chi^{2}$-distributed. When the cross-sectional variables are $s$-dependent (i.e., at most $s$ elements are dependent), the test still works well in the case with $p>n$. Using a suitable function of the data, the norm-based test can be applied to the heavy-tailed time series. We next show that the sample rank autocorrelation function (Spearman's rank correlaion) of the $L_{1}-$norm of data is asymptotically normal and the norm-based rank test statistic is asymptotically $\chi^{2}$-distributed. Surprisingly, the norm-based rank test is dimension-free, i.e. independent of $p$, and without requiring any moment condition of the data or the covariance structure condition as required in the literature. Two standardized norm-based tests are further discussed. Simulation results show that these test statistics have satisfactory sizes and are very powerful even for small $n$ and large $p$. A real data example is given.