Computes the auto-distance covariance function of a univariate time series. It also computes the unbiased estimator of squared auto-distance covariance.
ADCV(x, MaxLag = 15, unbiased = FALSE)
numeric vector or univariate time series.
maximum lag order at which to calculate the ADCV
. Default is 15.
logical value. If unbiased = TRUE, the unbiased estimator of squared auto-distance covariance is returned. Default value is FALSE.
Returns a vector, whose length is determined by MaxLag
, and contains the biased estimator of ADCV
or the unbiased estimator of squared ADCV
.
Szekely et al. (2007) recently proposed distance covariance function between two random vectors. Zhou (2012) extended this measure of dependence to a time series framework by calling it auto-distance covariance function.
ADCV
computes the sample auto-distance covariance function, \(V_X(\cdot)\), between \(\{X_t\}\) and \(\{X_{t+j}\}\).
Formal definition of \(V_X(\cdot)\) can be found in Zhou (2012) and Fokianos and Pitsillou (2016).
The empirical auto-distance covariance function, \(\hat{V}_X(\cdot)\), is the non-negative square root defined by $$ \hat{V}_X^2(j) = \frac{1}{(n-j)^2}\sum_{r,l=1+j}^{n}{A_{rl}B_{rl}}, \quad 0 \leq j \leq (n-1)$$
and \(\hat{V}_X^2(j) = \hat{V}_X^2(-j)\), for \(-(n-1) \leq j < 0\), where \(A=A_{rl}\) and \(B=B_{rl}\) are Euclidean distances with elements given by $$ A_{rl} = a_{rl} - \bar{a}_{r.} - \bar{a}_{.l} + \bar{a}_{..} $$ with \(a_{rl}=|X_r-X_l|\), \(\bar{a}_{r.}=\Bigl(\sum_{l=1+j}^{n}{a_{rl}}\Bigr)/(n-j)\), \(\bar{a}_{.l}=\Bigl(\sum_{r=1+j}^{n}{a_{rl}}\Bigr)/(n-j)\) , \(\bar{a}_{..}=\Bigl(\sum_{r,l=1+j}^{n}{a_{rl}}\Bigr)/(n-j)^2\). \(B_{rl}\) is given analogously based on \(b_{rl}=|Y_r-Y_l|\), where \(Y_t=X_{t+j}\). \(X_t\) and \(X_{t+j}\) are independent if and only if \(V_X^2(j)=0\). See Fokianos and Pitsillou (2016) for more information on theoretical properties of \(V_X^2(\cdot)\) including consistency.
If unbiased = TRUE, ADCV
returns the unbiased estimator of squared auto-distance covariance function, \(\tilde{V}_X^2(j)\), proposed by Szekely and Rizzo (2014).
In the context of time series data, this is given by
$$ \tilde{V}_X^2(j) = \frac{1}{(n-j)(n-j-3)}\sum_{r\neq l}{\tilde{A}_{rl}\tilde{B}_{rl}},
$$
for \(n > 3\), where \(\tilde{A}_{rl}\) is the \((r,l)\) element of the so-called U-centered matrix \(\tilde{A}\), defined by
$$ \tilde{A}_{rl} = \frac{1}{n-j-2}\sum_{t=1+j}^{n}{a_{rt}}-\frac{1}{n-j-2}\sum_{s=1+j}^{n}{a_{sl}+\frac{1}{(n-j-1)(n-j-2)}\sum_{t,s=1+j}^{n}{a_{ts}}}, \quad i \neq j,
$$
with zero diagonal.
mADCV
gives the auto-distance covariance function of a multivariate time series.
Fokianos K. and M. Pitsillou (2016). Consistent testing for pairwise dependence in time series. Technometrics, http://dx.doi.org/10.1080/00401706.2016.1156024.
Szekely, G. J. and M. L. Rizzo (2014). Partial distance correlation with methods for dissimilarities. The Annals of Statistics \(\textbf{42}\), 2382-2412, dx.doi.org/10.1214/14-AOS1255.
Szekely, G. J., M. L. Rizzo and N. K. Bakirov (2007). Measuring and testing dependence by correlation of distances. The Annals of Statistics \(\textbf{35}\), 2769-2794, http://dx.doi.org/10.1214/009053607000000505.
Zhou, Z. (2012). Measuring nonlinear dependence in time series, a distance correlation approach. Journal of Time Series Analysis \(\textbf{33}\), 438-457, http://dx.doi.org/10.1111/j.1467-9892.2011.00780.x.
# NOT RUN {
x <- rnorm(500)
ADCV(x,18)
ADCV(BJsales,25)
# }
Run the code above in your browser using DataLab