# zbMATH — the first resource for mathematics

Linear and quadratic serial rank tests for randomness against serial dependence. (English) Zbl 0661.62091
The $$ARMA(p_ 1,p_ 2)$$ model $$X_ t-n^{-1/2}\sum^{p_ 1}_{i=1}a_ i X_{t-i}=\epsilon_ t+n^{-1/2}\sum^{p_ 2}_{i=1}b_ i \epsilon_{t-i}$$ is considered, where $$\underset \tilde{} a=(a_ 1,...,a_{p_ 1})'$$ and $$\underset \tilde{} b=(b_ 1,...,b_{p_ 2})'$$ are parameters and $$\epsilon_ t$$ is white noise with density f. $$\underset \tilde{} X^{(n)}=(X_ 1^{(n)},...,X_ n^{(n)})'$$ is the vector of $$observations.$$
H$${}_ 0^{(n)}$$ is the family of distributions of all i.i.d. n-tuples $$\underset \tilde{} X^{(n)}$$, and $$H^{(n)}_{\underset \tilde{} d,f}$$ are the ARMA alternatives specified by a density f and a vector $$\underset \tilde{} d=(d_ 1,...,d_ p)'$$ such that $$d_ i=a_ i+b_ i$$, $$p=\max (p_ 1,p_ 2)$$. $$H_ 0^{(n)}$$ and $$H^{(n)}_{d,f}$$ are contiguous, and the paper considers testing the null hypothesis of randomness against the ARMA alternatives. Let $$\underset \tilde{} R^{(n)}=(R_ 1^{(n)},...,R_ n^{(n)})'$$ be the vector of ranks of $$\underset \tilde{} X^{(n)}$$, and $S^{(n)}=(n-p)^{-1}\sum^{n}_{t=p+1}a_ n(R_ t^{(n)},...,R^{(n)}_{t-p})$ a linear serial rank statistic defined in terms of a score function $$a_ n$$. If $$\underset \tilde{} S^{(n)}$$ is a vector of such statistics and $$\underset \tilde{} m^{(n)}$$ its mean under $$H_ 0^{(n)}$$, $$n^{1/2}(\underset \tilde{} S^{(n)}-\underset \tilde{} m^{(n)})$$ could be used as a test statistic, and would be asymptotically N($$\underset \tilde{} O,\underset \tilde{} V)$$. Due to the difficulties with this procedure, the quadratic test statistic $n(\underset \tilde{} S^{(n)}-\underset \tilde{} m^{(n)})' V^{-2}(\underset \tilde{} S^{(n)}- \underset \tilde{} m^{(n)})$ is considered: it is asymptotically $$\chi^ 2$$, resp. noncentral $$\chi^ 2$$, under the two hypotheses. A version of this procedure is proved to be asymptotically most powerful at level $$\alpha$$ when $$\underset \tilde{} d$$ and f are specified, and asymptotically maximum most powerful with $$\underset \tilde{} d$$ unspecified, f specified.
For various choices of f these results are related to rank “portmanteau” statistics: van der Warden, Wilcoxon, Laplace, Spearman and Box-Pierce. A final section includes questions unanswered by this approach.
Reviewer: R.Mentz

##### MSC:
 62G10 Nonparametric hypothesis testing 62M10 Time series, auto-correlation, regression, etc. in statistics (GARCH)
Full Text:
##### References:
 [1] Beran R., Ann. Statist. 3 pp 402– (1975) [2] Bhattacharyya G. K., Handbook of Statistics 4 pp 89– (1984) [3] DOI: 10.2307/2284333 · Zbl 0224.62041 · doi:10.2307/2284333 [4] DOI: 10.1080/03610928608829288 · Zbl 0602.62075 · doi:10.1080/03610928608829288 [5] Govindarajulu Z., Time Series Analysis: Theory and Practice 4 (1983) [6] DOI: 10.1214/aoms/1177704476 · Zbl 0133.42001 · doi:10.1214/aoms/1177704476 [7] Hajek J., Theory of Rank Tests (1967) · Zbl 0162.50503 [8] DOI: 10.1214/aos/1176349662 · Zbl 0584.62064 · doi:10.1214/aos/1176349662 [9] Hallin M., C. R. Acad. Sci., Paris 301 pp 935– (1985) [10] M. Hallin, J.Fr. Ingenbleek, and M. L. Puri (1986 ) Linear and quadratic serial rank tests for randomness against serial dependence. Technical Report, Department of Mathematics, Indiana University, Bloomington. · Zbl 0661.62091 [11] Lecam L., Univ. Calif. Publ. Stat. 3 pp 37– (1960) [12] Roussas G. G., Contiguity of Probability Measures: Some Applications in Statistics (1972) · Zbl 0265.60003 · doi:10.1017/CBO9780511804373 [13] DOI: 10.1016/0047-259X(85)90051-X · Zbl 0563.62065 · doi:10.1016/0047-259X(85)90051-X [14] Wald A., Ann. Math. Statist. 14 pp 378– (1943) [15] DOI: 10.1007/BF00532676 · Zbl 0314.60028 · doi:10.1007/BF00532676
This reference list is based on information provided by the publisher or from digital mathematics libraries. Its items are heuristically matched to zbMATH identifiers and may contain data conversion errors. It attempts to reflect the references listed in the original paper as accurately as possible without claiming the completeness or perfect precision of the matching.