Abstract
There has been an increasing interest recently in the analysis of financial data that arrives at irregular intervals. An important class of models is the autoregressive Conditional Duration (ACD) model introduced by Engle and Russell (Econometrica 66:1127–1162, 1998, [22]) and its various generalizations. These models have been used to describe duration clustering for financial data such as the arrival times of trades and price changes. However, relatively few evaluation procedures for the adequacy of ACD models are currently available in the literature. Given its simplicity, a commonly used diagnostic test is the Box-Pierce/Ljung-Box statistic adapted to the estimated standardized residuals of ACD models, but its asymptotic distribution is not the standard one due to parameter estimation uncertainty. In this paper we propose a test for duration clustering and a test for the adequacy of ACD models using wavelet methods. The first test exploits the one-sided nature of duration clustering. An ACD process is positively autocorrelated at all lags, resulting in a spectral mode at frequency zero. In particular, it has a spectral peak at zero when duration clustering is persistent or when duration clustering is small at each individual lag but carries over a long distributional lag. As a joint time-frequency decomposition method, wavelets can effectively capture spectral peaks and thus are expected to be powerful. Our second test checks the adequacy of an ACD model by using a wavelet-based spectral density of the estimated standardized residuals over the whole frequency. Unlike the Box-Pierce/Ljung-Box tests, the proposed diagnostic test has a convenient asymptotic “nuisance parameter-free” property—parameter estimation uncertainty has no impact on the asymptotic distribution of the test statistic. Moreover, it can check a wide range of alternatives and is powerful when the spectrum of the standardized duration residuals is nonsmooth, as can arise from neglected persistent duration clustering, seasonality, calender effects and business cycles. For each of the two new tests, we propose and justify a suitable data-driven method to choose the finest scale—the smoothing parameter in wavelet estimation. This makes the methods fully operational in practice. We present a simulation study, illustrating the merits of the wavelet-based procedures. An application with tick-by-tick trading data of Alcoa stock is presented.
Access this chapter
Tax calculation will be finalised at checkout
Purchases are for personal use only
References
Admati, A. R., & Pfleiderer, P. (1988). A theory of intraday patterns: Volume and price variability. Review of Financial Studies, 1, 3–40.
Andrews, D. W. K. (1991). Heteroskedasticity and autocorrelation consistent covariance matrix estimation. Econometrica, 59, 817–858.
Andrews, D. W. K. (2001). Testing when a parameter is on the boundary of the maintained hypothesis. Econometrica, 69, 683–734.
Bauwens, L., & Giot, P. (2000). The logarithmic ACD model: An application to the bid-ask quote process of three NYSE stocks. Annals of Economics and Statistics, 60, 117–149.
Bauwens, L., & Giot, P. (2001). Econometric modelling of stock market intraday activity. Advances studies in theoretical and applied econometrics. Boston: Kluwer.
Bauwens, L., & Giot, P. (2003). Asymmetric ACD Models: Introducing price information in ACD models. Empir. Econ., 28, 709–731.
Beltrao, K., & Bloomfield, P. (1987). Determining the bandwidth of a kernel spectrum estimate. Journal of Time Series Analysis, 8, 21–38.
Bera, A. K., & Higgins, M. L. (1992). A test for conditional heteroskedasticity in time series models. Journal of Time Series Analysis, 13, 501–519.
Bizer, D. S., & Durlauf, S. N. (1990). Testing the positive theory of government finance. Journal of Monetary Economics, 26, 123–141.
Box, G. E. P., & Pierce, D. (1970). Distribution of residual autocorrelations in autoregressive integrated moving average time series models. Journal of the American Statistical Association, 65, 1509–1526.
Bühlmann, P. (1996). Locally adaptive lag-window spectral estimation. Journal of Time Series Analysis, 17, 247–270.
Daubechies, I. (1992). Ten lectures on wavelets. In CBS-NSF regional conferences in applied mathematics (Vol. 61). Philadelphia: Society for Industrial and Applied Mathematics.
Drost, F. C., & Werker, B. J. M. (2000). Efficient estimation in semiparametric time series: The ACD model. Working paper, Tilburg University.
Drost, F. C., & Werker, B. J. M. (2004). Semiparametric duration models. Journal of Business & Economic Statistics, 22, 40–50.
Duchesne, P. (2006a). On testing for serial correlation with a wavelet-based spectral density estimator in multivariate time series. Econometric Theory, 22, 633–676.
Duchesne, P. (2006b). Testing for multivariate autoregressive conditional heteroskedasticity using wavelets. Computational Statistics & Data Analysis, 51, 2142–2163.
Duchesne, P., Li, L., & Vandermeerschen, J. (2010). On testing for serial correlation of unknown form using wavelet thresholding. Computational Statistics & Data Analysis, 54, 2512–2531.
Duchesne, P., & Pacurar, M. (2008). Evaluating financial time series models for irregularly spaced data: A spectral density approach. Computers & Operations Research (Special Issue: Applications of OR in Finance), 35, 130–155.
Easley, D., & O’Hara, M. (1992). Time and the process of security price adjustment. The Journal of Finance, 19, 69–90.
Engle, R. F. (2000). The econometrics of ultra-high frequency data. Econometrica, 68, 1–22.
Engle, R. F., & Russell, J. R. (1997). Forecasting the frequency of changes in quoted foreign exchange prices with the autoregressive conditional duration model. Journal of Empirical Finance, 4, 187–212.
Engle, R. F., & Russell, J. R. (1998). Autoregressive conditional duration: A new model for irregularly spaced transaction data. Econometrica, 66, 1127–1162.
Escanciano, J. C. (2006). Goodness-of-fit tests for linear and nonlinear time series models. Journal of the American Statistical Association, 101, 531–541.
Fan, Y., & Gençay, R. (2010). Unit root tests with wavelets. Econometric Theory, 26, 1305–1331.
Francq, C., Roy, R., & Zakoïan, J.-M. (2005). Diagnostic checking in ARMA models with uncorrelated errors. Journal of the American Statistical Association, 100, 532–544.
Francq, C., & Zakoïan, J.-M. (2009). Testing the nullity of GARCH coefficients: Correction of the standard tests and relative efficiency comparisons. Journal of the American Statistical Association, 104, 313–324.
Fernandes, M., & Grammig, J. (2005). Non-parametric specification tests for conditional duration models. Journal of Econometrics, 127, 35–68.
Gao, H. (1993). Wavelet estimation of spectral densities in time series analysis. Ph.D Dissertation, Department of Statistics, University of California, Berkeley.
Gençay, R., & Signori, D. (2012). Multi-scale tests for serial correlation. Technical report, Department of Economics, Simon Fraser University.
Gençay, R., Yazgan, E., & Ozkan, H. (2012). A test of structural change of unknown location with wavelets. Technical report, Department of Economics, Simon Fraser University.
Ghysels, E., & Jasiak, J. (1994). Stochastic volatility and time deformation: an application to trading volume and leverage effects. Working paper, C.R.D.E., Université de Montréal.
Ghysels, E., & Jasiak, J. (1998). Long-term dependence in trading. Working paper, Dept. of Economics, Penn State University and York University.
Ghysels, E., & Jasiak, J. (1998). GARCH for irregularly spaced financial data: The ACD-GARCH model. Studies in Nonlinear Dynamics and Econometrics, 2, 133–149.
Grammig, J., Hujer, R., Kokot, S., & Maurer, K.-O. (1998). Modeling the Deutsche Telekom IPO using a new ACD specification, an application of the Burr-ACD model using high frequency IBIS data. Working paper, Department of Economics, Johann Wolfgang Goethe-University of Frankfurt.
Granger, C. (1966). The typical spectral shape of an economic variable. Econometrica, 34, 150–161.
Granger, C. W. J., & Newbold, P. (1986). Forecasting economic time series (2nd ed.). New York: Academic Press.
Hall, P., & Heyde, C. C. (1980). Martingale limit theory and its application. New York: Academic Press.
Hannan, E. (1970). Multiple time series. New York: Wiley.
Hautsch, N. (2006). Testing the conditional mean function of autoregressive conditional duration models. Working Paper, University of Copenhagen.
Hernandez, E., & Weiss, G. (1996). A first course on wavelets. New York: CRC Press.
Higgins, M. L. (2000). Testing for autoregressive conditional duration. Presented at World Congress of Econometric Society, Seattle.
Hong, Y. (2001). Wavelet-based estimation for heteroskedasticity and autocorrelation consistent variance-covariance matrices. Working paper, Department of Economics and Department of Statistical Science, Cornell University.
Hong, Y., & Kao, C. (2004). Wavelet-based testing for serial correlation of unknown form in panel models. Econometrica, 72, 1519–1563.
Hong, Y., & Lee, J. (2001). One-sided testing for ARCH effect using wavelets. Econometric Theory, 17, 1051–1081.
Hong, Y., & Lee, Y.-J. (2011). Detecting misspecifications in autoregressive conditional duration models and non-negative time-series processes. Journal of Time Series Analysis, 32, 1–32.
Jasiak, J. (1999). Persistence in intertrade durations. Finance, 19, 166–195.
Jensen, M. J. (2000). An alternative maximum likelihood estimator of long memory processes using compactly supported wavelets. Journal of Economic Dynamics and Control, 24, 361–387.
Kyle, A. (1985). Continuous time auctions and insider trading. Econometrica, 53, 1315–1336.
Lee, J. H. H., & King, M. L. (1993). A locally most mean powerful based score test for ARCH and GARCH regression disturbances. Journal of Business & Economic Statistics, 11, 17–27 (Correction 1994, 12, p. 139).
Lee, J., & Hong, Y. (2001). Testing for serial correlation of unknown form using wavelet methods. Econometric Theory, 17, 386–423.
Li, L., Yao, S., & Duchesne, P. (2014). On wavelet-based testing for serial correlation of unknown form using Fan’s adaptive Neyman method. Computational Statistics & Data Analysis, 70, 308–327.
Li, W. K. (2004). Diagnostic checks in time series. New York: Chapman & Hall/CRC.
Li, W. K., & Mak, T. K. (1994). On the square residual autocorrelations in non-linear time series with conditional heteroskedasticity. Journal of Time Series Analysis, 15, 627–636.
Li, W. K., & Yu, L. H. (2003). On the residual autocorrelation of the autoregressive conditional duration model. Economics Letters, 79, 169–175.
Mallat, S. G. (1989). A theory for multiresolution signal decomposition: The wavelet representation. IEEE Transactions on Pattern Analysis and Machine Intelligence, 11, 674–693.
Meitz, M., & Teräsvirta, T. (2006). Evaluating models of autoregressive conditional duration. Journal of Business & Economic Statistics, 24, 104–124.
Neumann, M. H. (1996). Spectral density estimation via nonlinear wavelet methods for stationary non-Gaussian time series. Journal of Time Series Analysis, 17, 601–633.
Newey, W. K., & West, K. D. (1994). Automatic lag selection in covariance matrix estimation. Review of Financial Studies, 61, 631–653.
Pacurar, M. (2008). Autoregressive conditional duration models in finance: A survey of the theoretical and empirical literature. Journal of Economic Surveys, 22, 711–751.
Percival, D. B., & Walden, A. T. (2000). Wavelet methods for time series analysis, Cambridge Series in Statistical and Probabilistic Mathematics. Cambridge: Cambridge University Press.
Priestley, M. B. (1981). Spectral analysis and time series. London: Academic Press.
Priestley, M. B. (1996). Wavelets and time-dependent spectral analysis. Journal of Time Series Analysis, 17, 85–103.
Ramsey, J. (1999). The contribution of wavelets to the analysis of economic and financial data. Philosophical Transactions of the Royal Society A: Mathematical, Physical and Engineering, 537, 2593–2606.
Robinson, P. M. (1991). Automatic frequency domain inference on semiparametric and nonparametric models. Econometrica, 59, 1329–1363.
Stock, J. (1988). Estimating continuous time processes subject to time deformation. Journal of the American Statistical Association, 83, 77–85.
Tauchen, G., & Pitts, M. (1983). The price variability-volume relationship on speculative markets. Econometrica, 51, 485–505.
Tsay, R. S. (2005). Analysis of financial time series (2nd ed.). New York: Wiley.
Tsay, R. S. (2013). An introduction to analysis of financial data with R. New York: Wiley.
Vidakovic, B. (1999). Statistical modeling by wavelets. New York: Wiley.
Wang, Y. (1995). Jump and sharp cusp detection by wavelets. Biometrika, 82, 385–397.
Watson, N. W. (1993). Measures of fit for calibrated models. Journal of Political Economy, 101, 1011–1041.
Xue, Y., Gençay, R., & Fagan, S. (2010). Jump detection with wavelets. Technical report, Department of Economics, Simon Fraser University.
Zhang, M. Y., Russell, J. R., & Tsay, R. S. (2001). A nonlinear autoregressive conditional duration model with applications to financial transaction data. Journal of Econometrics, 104, 179–207.
Zhu, K., & Li, W. K. (2015). A bootstrapped spectral test for adequacy in weak ARMA models. Journal of Econometrics, 187, 113–130.
Acknowledgements
The authors would like to thank W. K. Li, David Stanford, Hao Yu, and two referees for constructive suggestions, which led to an improved paper. Funding in partial support of this work was provided by the Natural Science and Engineering Research Council of Canada.
Author information
Authors and Affiliations
Corresponding author
Editor information
Editors and Affiliations
Appendix
Appendix
To prove Theorems 1–3, we first state some useful lemmas.
Lemma 1
Define \(d_{J}(h)\equiv \sum _{j=0}^{J}\lambda (2\pi h/2^{j}),\) \(h,J\in \mathbb {Z}\), where \(\lambda (z)\) is as in (8). Then
-
(i)
\(d_{J}(0)=0\) and \(d_{J}(-h)=d_{J}(h)\) for all \(h,J\in \mathbb {Z},J>0\);
-
(ii)
\(|d_{J}(h)|\le C<\infty \) uniformly in \(h,J\in \mathbb {Z},J>0\);
-
(iii)
For any given \(h\in \mathbb {Z},h\ne 0\), \(d_{J}(h)\rightarrow 1\) as \(J\rightarrow \infty \);
-
(iv)
For any given \(r \ge 1\), \(\sum _{h=1}^{n-1}|d_{J}(h)|^{r}=O(2^{J})\) as \(J,n\rightarrow \infty \).
Lemma 2
Let \(V_{n}(J)\) and \(V_{0}\) be defined as in Theorem 1. Suppose \(J\rightarrow \infty , 2^{J}/n\rightarrow 0\). Then \(2^{-J}V_{n}(J)\rightarrow V_{0}\), where \(V_{0}=\int _{0}^{2\pi }|\varGamma (z)|^{2}dz\), with \(\varGamma (z)=\sum _{-\infty }^{\infty }\hat{\psi }(z+2\pi m)\).
Proof
For the proofs of Lemmas 1 and 2, see [42, Proof of Lemma A.1] and [44, Proof of Lemma A.2]. \(\Box \)
Proof
(Theorem 1) The model under the null hypothesis is \(D_t \equiv \beta _0\), \(X_t = \beta _0 \varepsilon _t\), \( E(\varepsilon _t)=1\), \(\{ \varepsilon _t \}\) an iid process. We write \(\bar{R}_X(h)= n^{-1}\sum _{t=|h|+1}^n (X_t/\bar{X}-1)(X_{t-|h|}/\bar{X}-1)\). Alternatively, \(\hat{\rho }_X(h)= \bar{R}_X(h)/\bar{R}_X(0)\). Under the null hypothesis, \( \bar{R}_X(h) = \bar{R}_{\varepsilon }(h)\) where \(\bar{R}_{\varepsilon }(h)\) is defined similarly as \(\bar{R}_X(h)\). We define \(R_{\varepsilon }(h)=\text {cov}(\varepsilon _t,\varepsilon _{t-h})\). Let \(u_t=\varepsilon _t-1\). We define \(\tilde{R}_{\varepsilon }(h)=n^{-1}\sum _{t=|h|+1}^n u_t u_{t-h}\) and \(\tilde{\rho }_{\varepsilon }(h) = \tilde{R}_{\varepsilon }(h)/R_{\varepsilon }(0)\). Define
where \(\tilde{\alpha }_{jk} \equiv (2\pi )^{-1/2}\sum _{h=1-n}^{n-1} \tilde{ \rho }_{\varepsilon }(h)\hat{\varPsi }_{jk}^{*}(h)\), \(\hat{\varPsi } _{jk}(h)\equiv (2\pi )^{-1/2}\int _{-\pi }^{\pi } \varPsi _{jk}(\omega )e^{-ih\omega }d\omega \).
Writing \(\hat{f}_X(0)-(2\pi )^{-1}=[\hat{f}_X(0)-\tilde{f}_{\varepsilon }(0)]+ [\tilde{f}_{\varepsilon }(0)-(2\pi )^{-1}],\) we shall prove Theorem 1 by showing Theorems 7–8 below.
Theorem 7
\([ V_{n}(J) ]^{-1/2}n^{1/2}[\hat{f}_X(0)-\tilde{f}_{\varepsilon }(0)]\rightarrow ^{p}0.\)
Theorem 8
\([ V_{n}(J) ]^{-1/2}n^{1/2}\pi [\tilde{f}_{\varepsilon }(0)-(2\pi )^{-1}]\rightarrow ^{d} N(0,1).\)
Proof
(Theorem 7) We use the following representations (see [44]):
We obtain \(\pi [\hat{f}_X(0)-\tilde{f}_{\varepsilon }(0)]=\sum _{h=1}^{n-1}d_{J}(h) [\hat{\rho }_X(h)-\tilde{\rho }_{\varepsilon }(h)]\). Because \(\bar{R}_X(0)-R_{\varepsilon }(0)=O_{P}(n^{-1/2})\) given Assumption 1, it suffices to show
We shall show (21) for the case \(J\rightarrow \infty \), where \(2^{-J}V_{n}(J)\rightarrow V_{0}\) by Lemma 2. The proof for fixed J is similar, with \(V_{n}(J)\rightarrow V_{0}(J)\), where \(V_{0}(J)\) is defined in Lemma 2.
Straightforward algebra yields \(\hat{R}_X(h)-\tilde{R}_{\varepsilon }(h)= (\bar{\varepsilon }^{-1}-1)\hat{A}_{1}(h)+ (\bar{\varepsilon }^{-1}-1)\hat{A}_{2}(h)+ (\bar{\varepsilon }^{-1}-1)^2\hat{A}_{3}(h)\), where
We first consider \(\hat{A}_{1}(h)\). Note that \(E[\hat{A}_1(h)\hat{A}_1(m)] = O(n^{-1})\), \(\forall h,m\). Then expending the square, we show that \(E[\sum _{h=1}^{n-1}d_J(h)\hat{A}_1(h)]^2 = O(2^J/n + 2^{2J}/n)\). This shows that \(\sum _{h=1}^{n-1} d_J(h) \hat{A}_1(h) = O_P(2^J/n^{1/2})\). Proceeding similarly we show that \(\sum _{h=1}^{n-1} d_J(h) \hat{A}_2(h) = O_P(2^J/n^{1/2})\). We show also easily that
This completes the proof for Theorem 7. \(\Box \)
Proof
(Theorem 8) Put \(\hat{W}\equiv \sum _{h=1}^{n-1}d_{J}(h)\tilde{R}_{\varepsilon }(h)/R_{\varepsilon }(0).\) Write \(\hat{W}=n^{-1}\sum _{t=2}^{n}W_{t}\), where
Observe that \(\{W_{t},\mathscr {F}_{t-1}\}\) is an adapted martingale difference sequence, where \(\mathscr {F}_{t}\) is the sigma field consisting of all \(u_{s},s\le t.\) Thus, we obtain
By the martingale central limit theorem in [37, pp.10–11], \([ V_{n}(J) ]^{-1/2}n^{1/2}\hat{W}\rightarrow ^{d}N(0,1)\) if we can show
For space, we shall show the central limit theorem for \(\hat{W}\) for large J (i.e., \(J\rightarrow \infty ).\) The proof for fixed J is similar and simpler because \(d_{J}(h)\) is finite and summable.
We shall verify the first condition by showing \(2^{-2J}n^{-2}\sum _{t=2}^{n}E(W_{t}^{4}) \rightarrow 0\). Put \(\mu _{4} \equiv E(u_{t}^{4})\). By Assumption 1, we have
It follows from Lemma 2 that \(2^{-2J}n^{-2}\sum _{t=1}^{n}E(W_{t}^{4})=O(n^{-1}).\) This show (22).
Next, given Lemma 2, it suffices for expression (23) to establish the sufficient condition
By the definition of \(W_{t}\), we have
It follows that
Whence, it suffices to show \(2^{-2J}[{\text {var}}(\hat{A})+{\text {var}}(\hat{B})]\rightarrow 0\). First, noting that \(A_{t}\) is a weighted sum of independent zero-mean variables \(\{u_{t-h}^{2}-R_{\varepsilon }(0)\},\) we have \(E(A_{t}^{2})=[\mu _{4}-R^2_{\varepsilon }(0)]\sum _{h=1}^{t-1}d_{J}^{4}(h).\) It follows by Minkowski’s inequality and Lemma 1(iv) that
Next, we consider \({\text {var}}(\hat{B})\). For all \(t\ge s,\) we have
where \(\delta _{j,h}=1\) if \(h=j\) and \(\delta _{j,h}=0\) otherwise. It follows that
by Lemma 1(iv). Combining (24)–(26) yields \(2^{-2J}[{\text {var}}(\hat{A})+{\text {var}}(\hat{B})]=O(2^{-J}+2^{J}/n)\rightarrow 0\) given \(J\rightarrow \infty , 2^{J}/n\rightarrow 0.\) Thus, condition (23) holds. By [37, pp.10–11], \([ V_{n}(J) ]^{-1/2}n^{1/2}\hat{W}\rightarrow ^{d}N(0,1)\). This completes the proof of Theorem 8. \(\Box \)
Proof
(Theorem 2) We shall show for large J only; the proof for fixed J is similar. Here we explicitly denote \(\hat{f}_X(0;J) \) as the spectral estimator (20) with the finest scale J. Recalling the definition of \(\mathscr {E}(J)\), we have
Note that we have for any given constants \(C_0 >0\) and \(\varepsilon >0\),
We now study \(|d_{\hat{J}}(h)-d_J(h)|\). We show that
Note that given \(C_0 2^{J/2}|2^{\hat{J}}/2^J - 1| \le \varepsilon \), we have that
and the lower and upper bounds in the summation are now non stochastic. Since \(|\sum _{m=-\infty }^{\infty } \psi (2\pi h/2^j + 2\pi m)| \le C\), we have that
Since \(\sum _{h=-\infty }^{\infty }|\hat{\psi }(2 \pi h/2^j)| \le C 2^j\), then
A similar argument allows us to show that
We show that \(V_n(\hat{J})/V_n(J) \rightarrow ^p 1\). Note that
Since \(\sum _{h=1}^{n-1} d_J^2(h) = O(2^J)\), by (27) and (28), we have the announced result that \(V_n(\hat{J})/V_n(J) \rightarrow ^p 1\).
Because \(\mathscr {E}(J)=O_{P}(1)\) by Theorem 1 and since \(V_{n}(\hat{J})/V_{n}(J)\rightarrow ^{p} 1\), we have \(\mathscr {E}(\hat{J})-\mathscr {E}(J)\rightarrow ^{p} 0\) provided \([V_{n}(J)]^{-1/2} n^{1/2}\pi \{\hat{f}_{\hat{J}}(0)-\hat{f}_{J}(0)\} \rightarrow ^{p}0,\) which we shall show below. The asymptotic normality of \(\mathscr {E}(\hat{J})\) follows from a standard application of Slutsky’s theorem and Theorem 1.
Because \(V_{n}(J)=O(2^{J}),\) it suffices to show \(\hat{f}_{X}(0;\hat{J})-\hat{f}_{X}(0;J)=o_{P}(2^{J/2}/n^{1/2}).\) Write
Given \(|d_{\hat{J}}(h)-d_{J}(h)|\le \sum _{j=\min (\hat{J},J)}^{\max (\hat{J},J)}|\lambda (2\pi h/2^{j})|\), we have, under the null hypothesis, the following inequality
We obtain \(\hat{f}_{X}(0;\hat{J})-\hat{f}_{X}(0;J)=o_{P}(2^{J/2}/n^{1/2}).\) This completes the proof of Theorem 2. \(\Box \)
Proof
(Theorem 3) Recall \(\hat{R}_X(h)= n^{-1}\sum _{t=|h|+1}^n (X_t-\bar{X})(X_{t-|h|}-\bar{X})\) and \(\tilde{R}_X(h)= n^{-1}\sum _{t=|h|+1}^n (X_t-\mu )(X_{t-|h|}-\mu )\). We study \(\hat{f}_X(0;J) - f_X(0)\). Write
where
We can show that \(E[ \hat{f}_X(0;J) - \tilde{f}_X(0;J) ]^2 = O(n^{-2} + 2^{2J}/n^2)\), meaning that replacing \(\bar{X}\) by \(\mu \) has to impact asymptotically. For the second term in (29), we show that
From [38, p. 313], we have
where the function \(w_{n}(u,h,m)\) is defined in [38]. We write
where
It follows that
Following a reasoning similar to [42, proof of Theorem 4.2], we can show that
and \(\left| A_n \right| =O(2^{J/2}/n)=o(2^{J}/n)\). It follows that
and
We consider the bias term \(E[\tilde{f}_{X}(J)]-f_{X}(0)\) in (29). Using the definition of \(\tilde{f}_{X}(0;J)\) in (30), we can decompose
Following a reasoning similar to [42, proof of Theorem 4.2], we can show that
\(\left| B_{5n}\right| \le n^{-\min (1,q)}\sum _{h=1}^{n-1}l^{q}|R_{X}(h)|=O(n^{-\min (1,q)})\) and also that \(\left| B_{6n}\right| \le 2n^{-q}\sum _{h=n}^{\infty }h^{q}|R_{X}(h)|=o(n^{-q})\). The bias term is then
Now, combining (29), (35), (37) we obtain
The desired result follows by using \(2^{J+1}/n^{\frac{1}{2q+1}}\rightarrow c\). This completes the proof of Theorem 3. \(\Box \)
Proof
(Corollary 1) The result follows immediately from Theorem 2 because the conditions of Corollary 1 imply \(2^{\hat{J}}/2^{J}-1=o_{P}(n^{-1/2(2q+1)})=o_{P}(2^{-J/2}),\) where the nonstochastic finest scale J is given by \(2^{J+1}\equiv \max \{[q\lambda _{q}^{2}\alpha (q)n/D_{\psi }]^{1/(2q+1)},0\}.\) The latter satisfies the conditions of Theorem 2. \(\Box \)
To prove Theorems 4–6, we first state a useful lemma.
Lemma 3
Suppose Assumptions 1 and 2 hold, \(J\rightarrow \infty \), and \(2^{J}/n\rightarrow 0\). Define
where \(a_{J}(h,m)=2\pi \sum _{j=0}^{J}\sum _{k=1}^{2^{j}}\hat{\psi }_{jk}(2\pi h)\hat{\psi }_{jk}^{*}(2\pi m).\) Then
-
(i)
\(b_{J}(h,m)\) is real-valued, \(b_{J}(0,m)=b_{J}(h,0)=0\) and \(b_{J}(h,m)=b_{J}(m,h)\);
-
(ii)
\(\sum _{h=1}^{n-1}\sum _{m=1}^{n-1}h^{\nu }|b_{J}(h,m)|=O(2^{(1+\nu )J})\) for \(0\le \nu \le \frac{1}{2}\);
-
(iii)
\(\sum _{h=1}^{n-1}\{\sum _{m=1}^{n-1}|b_{J}(h,m)|\}^{2}=O(2^{J})\);
-
(iv)
\(\sum _{h_{1}=1}^{n-1}\sum _{h_{2}=1}^{n-1}\{\sum _{m=1}^{n-1}|b_{J}(h_{1},m)b_{J}(h_{2},m)|\}^{2}=O\{(J+1)2^{J})\);
-
(v)
\(\sum _{h=1}^{n-1}b_{J}(h,h)=(2^{J+1}-1) \{1\)+\(O((J+1)/2^{J}\)+\(2^{J(2\tau -1)}/n^{2\tau -1})\},\) where \(\tau \) is in Assumption 3;
-
(vi)
\(\sum _{h=1}^{n-1}\sum _{m=1}^{n-1}b_{J}^{2}(h,m)=2(2^{J+1}-1)\{1+o(1)\}\);
-
(vii)
\(\sup _{1\le h,m\le n-1}|b_{J}(h,m)|\le C(J+1)\);
-
(viii)
\(\sup _{1\le h\le n-1}\sum _{m=1}^{n-1}|b_{J}(h,m)|\le C(J+1)\).
Proof
(Lemma 3) See [50, Appendix B] for (i)–(vi) and [43, Lemma A.1] for (vii) and (viii). \(\Box \)
Proof
(Theorem 4) Let \(Z_{t}=\varepsilon _{t}-1\) be such that \(E(Z_{t})=0\). Under \(\mathbb {H}_{0}^{\mathscr {A}}\), \(\varepsilon _{t}=e_{t}\). We consider \(\bar{R}_{Z}(h)=n^{-1}\sum _{t=|h|+1}^{n}Z_{t}Z_{t-|h|}\) and \(\bar{\alpha }_{ejk}=\sum _{h=1-n}^{n-1}\bar{R}_{Z}(h)\hat{\psi }_{jk}^{*}(2\pi h)\). Let \(\bar{\mathscr {A}}(J)\) defined as \(\mathscr {A}(J)\) but using \(\bar{\alpha }_{ejk}\). Let \(\bar{\rho }_{Z}(h)=\bar{R}_{Z}(h)/\bar{R}_{Z}(0)\). Because \(\bar{\rho }_{Z}(-h)=\bar{\rho }_{Z}(h)\) and \(\bar{\alpha }_{ejk}\) is real-valued, we have
where the equality follows from re-indexing and the definition of \(b_{J}(h,m)\). We have \(\bar{R}_Z(0)-\sigma _Z^{2}=O_{P}(n^{-1/2})\), since under \(\mathbb {H}_{0}^{\mathscr {A}}\) we have that \(\{ e_t \}\) is iid. Also, we have
where the second term is of the indicated order of magnitude because
given \(E[\bar{R}_Z^{2}(h)] \le Cn^{-1}\) and Lemma 3(ii). We now focus on the first term in (38). We have,
where
In order to prove Theorem 4, we first state Proposition 1 that shows that \(\hat{C}_{n}\) represents the dominant term.
Proposition 1
Suppose Assumptions 1–3 hold, \(J\rightarrow \infty \), and \(2^{2J}/n\rightarrow 0\). Then
We now decompose \(\hat{C}_{n}\) into the terms with \(t-s>q\) and \(t-s\le q\), for some \(q\in \mathbb {Z}^{+}\):
Furthermore, we decompose
where \(\hat{D}_{5n}\) and \(\hat{D}_{6n}\) are the contributions from \(m>q\) and \(h>q\), respectively.
Proposition 2 shows that \(\hat{C}_{n}\) can be approximated arbitrarily well by \(\hat{U}_{n}\) under a proper condition on q.
Proposition 2
Suppose Assumptions 1–3 hold, \(J\rightarrow \infty ,2^{2J}/n\rightarrow 0\), and \(q\equiv q_{n}\rightarrow \infty ,q/2^{J}\rightarrow \infty ,q^{2}/n\rightarrow 0\). Then \(2^{-J/2}\hat{C}_{n}=2^{-J/2}\hat{U}_{n}+o_{P}(1).\)
It is much easier to show the asymptotic normality of \(\hat{U}_{n}\) than of \(\hat{C}_{n},\) because for \(\hat{U}_{n},\) \(\{Z_{t}Z_{t-h}\}\) and \(\{Z_{s}Z_{s-m}\}\) are independent given \(t-s>q\) and \(0<h,m\le q\).
Proposition 3
Suppose Assumptions 1–3 hold, and \(J\rightarrow \infty ,2^{2J}/n\rightarrow 0,q/2^{J}\rightarrow \infty , q^{2}/n\rightarrow 0\). Let \(\lambda _{n}^{2}=E(\hat{U}_{n}^{2})\). Then \(4(2^{J+1}-1)\sigma ^{8}/\lambda _{n}^{2}\rightarrow 1\), and \(\lambda _{n}^{-1}\hat{U}_{n}\rightarrow ^{d}N(0,1)\).
Propositions 1–3 and Slutsky’s Theorem imply \(\bar{\mathscr {A}}(J) \rightarrow ^{d} N(0,1)\). Propositions 4 and 5 show that parameter estimation does not have impact on the asymptotic distribution of the test statistic.
Proposition 4
\(n \sum _{j=0}^{J}\sum _{k=1}^{2^{j}} (\hat{\alpha }_{jk}-\bar{\alpha }_{jk})^{2}= O_P(2^J/n) + O_P(1).\)
Proposition 5
\(n \sum _{j=0}^{J}\sum _{k=1}^{2^{j}} (\hat{\alpha }_{jk}-\bar{\alpha }_{jk}) \bar{\alpha }_{jk} = o_P(2^{J/2}).\)
The proof of Theorem 4 will be completed provided Propositions 1–5 are shown. The proofs of Propositions 1–3 are very similar to the proofs of Propositions 1–3 in [50], for proving the asymptotic normality of a wavelet-based test statistic for serial correlation. These proofs are then omitted (but for the interested reader all the detailed proofs are available from the authors).
Proof
(Proposition 4) A standard Taylor’s expansion gives
where \(\bar{{\varvec{\theta }}}\) lies between \(\hat{ {\varvec{\theta }}}\) and \({\varvec{\theta }}_0\). We have
We write \(\hat{\alpha }_{jk} - \tilde{\alpha }_{jk} = \hat{B}_{1jk} + \hat{B} _{2jk}+\hat{B}_{3jk}\) where
Then \((\hat{\alpha }_{jk} - \tilde{\alpha }_{jk})^2 \le 4(\hat{B}_{1jk}^2 + \hat{B}_{2jk}^2+ \hat{B}_{3jk}^3)\) We first study the term involving \(\hat{B}_{1jk}\). We write
We have
We now study the term involving \(\hat{B}_{2jk}\). Let
We write \(\hat{A}_2(h) = \hat{A}_{21}(h) + \hat{A}_{22}(h)\), where
Then
We obtain
since
We now study the term involving \(\hat{B}_{3jk}\).
We write \(\hat{A}_3(h) = \hat{A}_{31}(h) + \hat{A}_{32}(h)\), where
We obtain
We write \(\hat{a}_{31}(h) = E[\hat{a}_{31}(h)] + \left\{ \hat{a}_{31}(h)-E[\hat{a}_{31}(h)]\right\} \). We have
Since we can interpret \(E\hat{a}_{31}(h) = {\text {cov}}(Z_{t-|h|}, X_t \frac{\partial }{\partial {\varvec{\theta }}} D_t^{-1}({\varvec{\theta }}_0))\) as a cross-correlation function, we have that
Also,
This shows Proposition 4. \(\Box \)
Remark 1
Proposition 4 is established under a general stationary process for \(\{ Z_t \}\), that is, the result is established without assuming the null hypothesis.
Proof
(Proposition 5) We write \((\hat{\alpha }_{jk}-\tilde{\alpha }_{jk})\tilde{\alpha }_{jk}=\hat{C}_{1jk}+\hat{C}_{2jk}+\hat{C}_{3jk}\), where
By the Cauchy–Schwarz inequality and the fact that \(n\sum _{j=0}^{J}\sum _{k=1}^{2^{j}}\tilde{\alpha }_{jk}^{2}=O_{P}(2^{J})\) under the null hypothesis, we have that
Since \(n\sum _{j=0}^{J}\sum _{k=1}^{2^{j}}B_{2jk}^{2}=O_{P}(2^{J}/n)\), we have that
We write
We have
Since
Then
We have
This completes the proof of Proposition 5 and so Theorem 4. \(\Box \)
Proof
(Theorem 5) We write
where \(C_{n}(J)=2^{J+1}-1\), \(D_{n}(J)=4(2^{J+1}-1)\). Given \(\mathscr {A}(J)=O_{P}(1)\) by Theorem 4, it suffices for \(\mathscr {A}(\hat{J})-\mathscr {A}(J)\rightarrow ^p 0\) and \(\mathscr {A}(\hat{J}) \rightarrow ^d N(0,1)\) to establish
We first show (i). Decompose
For the first term in (45), we write
By Proposition 4, we have \([D_{n}(J)]^{-1/2}\hat{G}_{12}\rightarrow ^p 0\). For the first term in (46), we have for any given constants \(M>0\) and \(\varepsilon >0\),
For any given constants \(C_{0},\varepsilon >0,\) the second term in (47) vanishes to 0 as \(n \rightarrow \infty \) given \(2^{J/2}|2^{\hat{J}}/2^{J}-1|\rightarrow ^{p} 0.\) For the first term, given \(C_{0}2^{J/2}|2^{\hat{J}}/2^{J}-1|\le \varepsilon \), we have for n sufficiently large,
by Proposition 4. Therefore, we have
Next, we consider \(\hat{G}_{2}\) in (45). We write
Since for n sufficiently large,
given Assumption 3 and
Cf. [61, (6.19), p.392]. Therefore
Then \([D_{n}(J)]^{-1/2}\left\{ \hat{G}_{2}-[C(\hat{J})-C(J)]\right\} \rightarrow ^p 0\). Note that \([D_{n}(J)]^{-1/2}[C_{n}(\hat{J})-C_{n}(J)]=o_{P}(1)\). Next, by the Cauchy-Schwarz inequality and (48), we have
Summarizing, we obtain result (i), that is
We now show (ii), that is \(D_{n}(\hat{J})/D_n(J)=1+o_{P}(1)\). Using the fact that \(2^{\hat{J}}/2^J = 1 + o_P(2^{-J/2})\) and \(J \rightarrow \infty \) such that \(2^J/n \rightarrow 0\), one shows easily that
This shows (ii). This completes the proof of Theorem 5. \(\Box \)
Proof
(Theorem 6) We first show \(Q(\hat{f},f)=Q(\tilde{f},f)+o_{P}(2^{J}/T+2^{-2qJ})\). Write
For the first term in (50), by Parseval’s identity, Proposition 4 (which can be shown to continues to hold given Assumptions 2–3 and 7–10; See Remark 1), and \(D_{n}(J)\propto O(2^{J+1})\), we have
as \(n\rightarrow \infty \). For the second term, we have \(\hat{Q}_{2}=o_{P}(2^{J}/n+2^{-2qJ})\) by the Cauchy-Schwarz inequality, (51) and the fact that \(Q(\tilde{f},f)=O_{P}(2^{J}/n+2^{-2qJ})\), which follows by Markov’s inequality and \(EQ(\tilde{f},f)=O(2^{J}/n+2^{-2qJ})\). The latter is to be shown below.
To compute \(E[Q(\tilde{f},f)],\) we write
We first consider the second term in (52). Put \(B(\omega ) \equiv \sum _{j=J+1}^{\infty }\sum _{k=1}^{2^{j}}\alpha _{jk}\varPsi _{jk}(\omega ) \). Then
We evaluate directly \(\int _{-\pi }^{\pi }B^{2}(\omega )d\omega \). Using the orthonormality of the wavelet basis, we have that
Replacing \(\alpha _{jk}=\sum _{h=-\infty }^{\infty }R_e(h)\hat{\psi }_{jk}(2 \pi h)\) and since \(\hat{\psi }_{jk}(2 \pi h)=e^{-i 2\pi h k /2^j} 2^{-j/2}\hat{\psi }(2\pi h/2^j)\),
where \(c_j(h,m) = 2^{-j} \sum _{k=1}^{2^j} e^{i 2\pi (m-h) k/2^j}\) is as in (49). By a change of variables,
We evaluate separately the case corresponding to \(r=0\) and \(r \ne 0\) in (54).
where \(f^{(q)}_e(\cdot )\) is defined in Sect. 4.3 and o(1) is uniform in \(\omega \in [-\pi ,\pi ].\) It follows that
It may be show that the term corresponding to \(r\ne 0\) is \(o(2^{-2qJ})\).
For the second term in (53), we have
given Lemma 3(vii) and \(\sum _{h=-\infty }^{\infty }|hR_e(h)|\le C\) as implied by Assumption 10.
Finally, we consider the variance factor in (52). We write
where the function \(\eta (l)\) satisfies
For more details see [61, p. 326]. Given Assumption 9 and Lemma 3(vii), we have \(|V_{2n}|\le C(J+1)n^{-1}\) and \(|V_{3n}|\le C(J+1)n^{-1}\). For the first term \(V_{1n}\), we can write
where we have used Lemma 3(v) for the first term, which corresponds to \(h=m \); the second term corresponds to \(h\ne m\) and it is \(O[(J+1)/T]\) given \(\sum _{h=-\infty }^{\infty }|R(h)|\le C\) and Lemma 3(v). It follows that as \(J\rightarrow \infty \)
Collecting (55)–(57) and \(J\rightarrow \infty \), we obtain
This shows the Theorem. \(\Box \)
Proof
(Corollary 2) The result follows immediately from Theorem 6 because Assumption 9 implies \(2^{\hat{J}}/2^{J}-1=o_{P}(T^{-1/2(2q+1)})=o_{P}(2^{-J/2})\), where the nonstochastic finest scale J is given by \(2^{J+1}\equiv \max \{[2\alpha \vartheta _{q}\zeta _{0}(q)T]^{1/(2q+1)},0\}.\) The latter satisfies the conditions of Theorem 6. \(\Box \)
Rights and permissions
Copyright information
© 2016 Springer Science+Business Media New York
About this chapter
Cite this chapter
Duchesne, P., Hong, Y. (2016). On Diagnostic Checking Autoregressive Conditional Duration Models with Wavelet-Based Spectral Density Estimators. In: Li, W., Stanford, D., Yu, H. (eds) Advances in Time Series Methods and Applications . Fields Institute Communications, vol 78. Springer, New York, NY. https://doi.org/10.1007/978-1-4939-6568-7_3
Download citation
DOI: https://doi.org/10.1007/978-1-4939-6568-7_3
Published:
Publisher Name: Springer, New York, NY
Print ISBN: 978-1-4939-6567-0
Online ISBN: 978-1-4939-6568-7
eBook Packages: Mathematics and StatisticsMathematics and Statistics (R0)