Abstract
This study aims to test for detecting a change point in the conditional quantile of general location-scale time series models. This issue is quite important in risk management because the conditional quantile is utilized to measure the value-at-risk or expected shortfall of financial assets. In this paper, we design two types of cumulative sum tests based on the conditional quantiles. Their limiting null distributions are derived under regularity conditions, together with consistency of the proposed tests under the alternative. Monte Carlo simulations demonstrate the good performance of the proposed tests in terms of both stability and power for various time series settings. A real data analysis using the daily returns of the Brent Oil futures also confirms the validity of the tests in real-world applications.
Similar content being viewed by others
References
Berkes, I., Horváth, L., Kokoszka, P. (2004). Testing for parameter constancy in GARCH\((p, q)\) models. Statistics & Probability Letters, 70, 263–273.
Billingsley, P. (1968). Convergence of probability measure. New York: Wiley.
Bloomfield, P., Steiger, W. L. (1983). Least absolute deviations: Theory, applications, and algorithms. Boston: Birkhäuser.
Bollerslev, T. (1986). Generalized autoregressive conditional heteroskedasticity. Journal of Econometrics, 31, 307–327.
Bougerol, P., Picard, N. (1992). Stationarity of GARCH processes and of some nonnegative time series. Journal of Econometrics, 52, 115–127.
Campbell, S. D. (2006). A review of backtesting and backtesting procedures. The Journal of Risk, 9, 1–17.
Chen, J., Gupta, A. K. (2012). Parametric statistical change point analysis with applications to genetics, medicine, and finance. New York: Wiley.
Christoffersen, P. F. (1998). Evaluating interval forecasts. International Economic Review, 39, 841–862.
Ciuperca, G. (2017). Real time change-point detection in a nonlinear quantile model. Sequential Analysis, 36, 87–110.
Csörgő, M., Horváth, L. (1997). Limit theorems in change-point analysis. New York: Wiley.
de Pooter, M., van Dijk, D. (2004). Testing for changes in volatility in heteroskedastic time series—A further examination. Econometric Institute Research Papers EI 2004-38, Erasmus University Rotterdam, Erasmus School of Economics (ESE), Econometric Institute.
Engle, R. F. (1982). Autoregressive conditional heteroscedasticity with estimates of the variance of United Kingdom inflation. Econometrica, 50, 987–1007.
Engle, R., Manganelli, S. (2004). CAViaR: Conditional autoregressive value at risk by regression quantiles. Journal of Business & Economic Statistics, 22, 367–381.
Fitzenberger, B., Koenker, R., Machado, J. (2013). Economic applications of quantile regression. Berlin, Heidelberg: Springer.
Francq, C., Zakoian, J.-M. (2004). Maximum likelihood estimation of pure GARCH and ARMA-GARCH processes. Bernoulli, 10, 605–637.
Gombay, E. (2008). Change detection in autoregressive time series. Journal of Multivariate Analysis, 99, 51–464.
Hall, P., Heyde, C. C. (1980). Martingale limit theory and its application. San Diego: Academic Press.
Inclán, C., Tiao, G. C. (1994). Use of cumulative sums of squares for retrospective detection of changes of variance. Journal of the American Statistical Association, 89, 913–923.
Kim, M., Lee, S. (2016). Nonlinear expectile regression with application to value-at-risk and expected shortfall estimation. Computational Statistics & Data Analysis, 94, 1–19.
Kim, S., Cho, S., Lee, S. (2000). On the cusum test for parameter changes in GARCH\((1,1)\) models. Communications in Statistics-Theory and Methods, 29, 445–462.
Kirch, C., Kamgaing, J. (2012). Testing for parameter stability in nonlinear autoregressive models. Journal of Time Series Analysis, 33, 365–385.
Koenker, R., Bassett, Jr., G. (1978). Regression quantiles. Econometrica: Journal of the Econometric Society, 46, 33–50.
Koenker, R., Zhao, Q. (1996). Conditional quantile estimation and inference for ARCH models. Econometric Theory, 12, 793–813.
Kupiec, P. (1995). Techniques for verifying the accuracy of risk measurement models. The Journal of Derivatives, 3, 73–84.
Lee, S. (2020). Location and scale-based CUSUM test with application to autoregressive models. Journal of Statistical Computation and Simulation, 90, 2309–2328.
Lee, S., Kim, C. K. (2022). Test for conditional quantile change in GARCH models. Journal of the Korean Statistical Society, 51, 480–499.
Lee, S., Noh, J. (2010). Value at risk forecasting based on quantile regression for GARCH models. The Korean Journal of Applied Statistics, 23, 669–681.
Lee, S., Noh, J. (2013). Quantile regression estimator for GARCH models. Scandinavian Journal of Statistics, 40, 2–20.
Lee, S., Ha, J., Na, O., Na, S. (2003). The CUSUM test for parameter change in time series models. Scandinavian Journal of Statistics, 30, 781–796.
Lee, S., Tokutsu, Y., Maekawa, K. (2004). The cusum test for parameter change in regression models with ARCH errors. Journal of Japan Statistical Society, 34, 173–188.
Meitz, M., Saikkonen, P. (2008). Ergodicity, mixing, and existence of moments of a class of Markov models with applications to GARCH and ACD models. Econometric Theory, 24, 1291–1320.
Noh, J., Lee, S. (2016). Quantile regression for location-scale time series models with conditional heteroscedasticity. Scandinavian Journal of Statistics, 43, 700–720.
Oh, H., Lee, S. (2018). On score vector-and residual-based CUSUM tests in ARMA-GARCH models. Statistical Methods & Applications, 27, 385–406.
Oh, H., Lee, S. (2019). Modified residual CUSUM test for location-scale time series models with heteroscedasticity. Annals of Institute of Statistical Mathematics, 71, 1059–1091.
Oka, T., Qu, Z. (2011). Estimating structural changes in regression quantiles. Journal of Econometrics, 162, 248–267.
Page, E. S. (1955). A test for a change in a parameter occurring at an unknown point. Biometrika, 42, 523–527.
Qu, Z. (2008). Testing for structural change in regression quantiles. Journal of Econometrics, 146, 170–184.
Straumann, D., Mikosch, T. (2006). Quasi-maximum-likelihood estimation in conditionally heteroscedastic time series: A stochastic recurrence equations approach. The Annals of Statistics, 34, 2449–2495.
Su, L., Xiao, Z. (2008). Testing for parameter stability in quantile regression models. Statistics & Probability Letters, 78, 2768–2775.
Weiss, A. (1991). Estimating nonlinear dynamic models using least absolute error estimation. Econometric Theory, 7, 46–68.
Xiao, Z., Koenker, R. (2009). Conditional quantile estimation for generalized autoregressive conditional heteroscedasticity models. Journal of the American Statistical Association, 104, 1696–1712.
Zhou, M., Wang, H. J., Tang, Y. (2015). Sequential change point detection in linear quantile regression models. Statistics & Probability Letters, 100, 98–103.
Acknowledgements
We sincerely thank the Editor, an AE, and anonymous reviewers for their careful reading and valuable comments that improve the quality of the paper. This research is supported by Basic Science Research Program through the National Research Foundation of Korea (NRF) funded by the Ministry of Science, ICT and Future Planning (No. 2021R1A2C1004009).
Author information
Authors and Affiliations
Corresponding author
Additional information
Publisher's Note
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
Supplementary Information
Below is the link to the electronic supplementary material.
Appendix
Appendix
In this section, we provide the proofs of some results in Sects. 2 and 3.
Proof of Theorem 1
It suffices to prove (9). Put
Noting that \(\psi _\tau (y_t-{\hat{q}}_t)=\tau -I(\eta _t \le \xi _0 +\varLambda _t({\hat{\vartheta _n}} ) )\), we can express
Owing to Proposition 1, for any \(\delta \in (0,1)\), there exists a (large enough) \(L>0\) such that \(P(\hat{\vartheta }_{n}\in \mathcal{N}_{L/\sqrt{n}})\ge 1-\delta\), where \({{{\mathcal {N}}}}_{L/\sqrt{n}}\) is a compact neighborhood of \(\vartheta _{0}\) with \(||\vartheta -\vartheta _{0}||\le L/\sqrt{n}\) for all \(\vartheta \in {{{\mathcal {N}}}}_{L/\sqrt{n}}\). Given any fixed \(\zeta >0\), we decompose \({{{\mathcal {N}}}}_{L/\sqrt{n}}\) into a finite number of subsets \(\mathcal{D}_1,\ldots ,{{{\mathcal {D}}}}_N\) for some \(N=N(\zeta )\ge 1\), with their diameters less than \(\zeta /\sqrt{n}\). We then choose points \(\vartheta _j\) from \({{{\mathcal {D}}}}_j\). Then, in case of \({\hat{\vartheta _n}} \in {{{\mathcal {D}}}}_j\), we have
with \(\varLambda _{tj}^{-}=\inf _{\vartheta \in {{{\mathcal {D}}}}_j} \varLambda _t (\vartheta )-\varLambda _t (\vartheta _j)\) and \(\varLambda _{tj}^{+}=\sup _{\vartheta \in {{{\mathcal {D}}}}_j} \varLambda _t (\vartheta )-\varLambda _t (\vartheta _j )\). Putting
and using (A5), (A6), and the mean value theorem, we can show that
with
which can be made arbitrarily small by taking a small \(\zeta\).
Next, putting
we set \(S_{kj}^+=\sum _{t=1}^{k}e_{tj}^+\) and \(S_{kj}^-=\sum _{t=1}^{k}e_{tj}^-\), Moreover, for \(B>0\), we set
with
Then, we get martingale arrays \(\{({{\tilde{S}}}_{kj}^+, {{{\mathcal {F}}}}_t); k=1,\ldots ,n\}\) and \(\{({{\tilde{S}}}_{kj}^-, {{{\mathcal {F}}}}_t); k=1,\ldots , n\}\), where \({{{\mathcal {F}}}}_t=\sigma (\eta _t, \eta _{t-1},\ldots )\).
Note that \(P ({{\tilde{e}}}_{tj}^+ \ne e_{tj}^+ \ \ \mathrm{for\ some}\ t=1,\ldots , n)\) can be made arbitrarily small by taking a large B, which can be seen using the mean value theorem, Hölder’s inequality, (A5) and (A6). Then, for any \(\zeta ^{'}, \lambda >0\), \(0<\delta <1\) and \(p>1\), using the sub-martingale inequality and taking a sufficiently large \(B:= B(\zeta )\), we can get
Moreover, using Rosenthal’s inequality (Hall and Heyde 1980), we get
and further,
which leads to \(\sum _{j=1}^N E{{\tilde{S}}}_{nj}^{+2p}=O(n^{p/2}+n )\). Then, in view of (14), we obtain
Analogously, we have \(\max _{1\le j\le N}\max _{1\le k\le n} |A_{kj}^-|=o_P (1).\) Hence, from (12) and (13), we have \(A_n= o_P (1)\).
Next, we handle \(B_n\). Set
Note that for \(k_n = n^{1/3}\), \(\frac{1}{\sqrt{n}}\max _{1\le k\le k_n} B_{nk}=o_P (1).\) Thus it suffices to show that
To this end, we use Taylor’s theorem to get
for some number \(\xi _t^*\) lying between \(\xi _0\) and \(\xi _0 +\varLambda _t ({\hat{\vartheta _n}})\). Then, using this, (A2), (A5), (A6), and our assumption on \(f_\eta ^{'}\), we can show that for some \(K>0\),
due to the ergodicity of \(\{y_t\}\), which entails \(B_n=o_P (1)\). As we can check \({\hat{\tau }}_{1}^2=\tau _1^2+o_P (1)\), we finally get \({\hat{T}}_{n}- T_{n}=o_P (1)\), which in turn implicates the weak convergence of \({\hat{T}}_{n}\) to \(\sup _{0\le s\le 1} | B^\circ (s)|\). \(\square\)
Proof of Theorem 4
For \(i=1, 2\), we set
Noting that \(\psi _\tau (y_{t1}-{\hat{q}}_{t1})=\tau -I(\eta _t \le \varLambda ^*_t ({\hat{\vartheta _n}} )+\xi _{t1} )\), we can express for \(k\le k_0\),
where
Then, following the lines similar to those used for verifying \(A_n=o_P (1)\) in the proof of Theorem 1, with \(\mathcal{N}_{L/\sqrt{n}}\), \(\varLambda _t (\vartheta )\), and \(\xi _0\), therein, replaced by \({{{\mathcal {N}}}}_L^*=\{\vartheta : ||\vartheta -\vartheta _0^*||\le L\}\), \(\varLambda _{t1}^{*}(\vartheta )\), and \(\xi _{t1}\), respectively, where L can be taken to be arbitrarily small, \({{{\mathcal {N}}}}_L^*\) and L play the role of the partition \({{{\mathcal {D}}}}_1\) and \(\zeta\) (namely, \({{{\mathcal {N}}}}_L^*=\mathcal{D}_1\), \(L=\zeta\), and \(N(\zeta )=1\)), and both \(A_{k1}^+\) and \(A_{k1}^-\) are accordingly reformulated with the denominator \(\sqrt{n}\) replaced by n, we can similarly verify that \(\max _{1\le k\le k_0} | I_{nk}|= o_P (1)\). Also, using (A2), (A5), (A6) (modified with \(g_{ti}, h_{ti}, \tilde{g}_{ti}, {{\tilde{h}}}_{ti}\)), (B1), the ergodicity, and the mean value theorem, we can readily check that \(\max _{1\le k\le k_0} | II_{nk}|= o_P (1)\). Subsequently, we get
uniformly in \(k\le k_0\). Similarly, additionally harnessing (11), we can verify
uniformly in \(k>k_0\). Therefore, we can have
due to the ergodicity and (B2), so that \({\hat{T}}_n \rightarrow \infty\) in probability as \({\hat{\tau _n}}=\tau _0+ o_P(1)\) for some \(\tau _0>0\).
Moreover, putting
we can see that \(\max _{0\le s\le 1}|\hat{{{\mathcal {L}}}}_n ([ns])- \mathcal{L}_n ([ns])|= o_P (1)\) by virtue of (15) and (16), where
Furthermore, using the ergodicity, we can easily check that \(\sup _{0\le s\le 1}|{{{\mathcal {L}}}}_n ([ns])-{{{\mathcal {L}}}}(s)|=o_P (1),\) where
which has a maximum value \(\kappa _0(1-\kappa _0)|d_1-d_2|\) uniquely at \(s=\kappa _0\). This with the argmax theorem yields \({\hat{k}}_n /n \rightarrow \kappa _0\) in probability, which validates the theorem. \(\square\)
On the issue regarding (B1) in Remark 4. For \(t\le k_0\), put
Then, we can express
with
We first verify
Due to (A5) and (A6), it suffices to show that
or equivalently, for \(\delta \in (0,1)\),
which we can check to hold via using the Cauchy–Schwarz inequality, provided
To verify (18), we partition \(\varTheta ^*\) into \({{{\mathcal {D}}}}_j\) with diameter less than \(\zeta\), \(j=1,\ldots , N=N(\zeta )\), and choose points \(\vartheta _j\) from \({{{\mathcal {D}}}}_j\). Then, in case of \(\vartheta \in {{{\mathcal {D}}}}_j\), we have
with \(\varLambda _{tj}^{\dagger (1)}=\inf _{\vartheta \in {{{\mathcal {D}}}}_j} \varLambda _t^\dagger (\vartheta )-\varLambda _t^\circ (\vartheta _j)\) and \(\varLambda _{tj}^{\dagger (2)}=\sup _{\vartheta \in {{{\mathcal {D}}}}_j} \varLambda _t^\dagger (\vartheta )-\varLambda _t^\circ (\vartheta _j )\). Also,
with \(\varLambda _{tj}^{\circ (1)}=\inf _{\vartheta \in {{{\mathcal {D}}}}_j} \varLambda _t^\circ (\vartheta )-\varLambda _t^\circ (\vartheta _j)\) and \(\varLambda _{tj}^{\circ (2)}=\sup _{\vartheta \in {{{\mathcal {D}}}}_j} \varLambda _t^\circ (\vartheta )-\varLambda _t^\circ (\vartheta _j )\).
By simple algebras using (A5) and (A6), it can be seen that
As \(\zeta\) can be taken to be arbitrarily small, \(C_n\) becomes \(o_P(1)\), which ensures (18) and thereby (17) as the quantity in (18) is no more than \(C_n\). Moreover, following the lines similar to the above, we can also check
and thus, due to (17),
Similarly, it can be shown that
then combining (19) and (20), we have
which validates the conclusion in Remark 4. Namely, (B1) holds true if \(\kappa _0 E\rho _\tau (y_{t1} - q_{t1}(\vartheta ))+ (1-\kappa _0)E \rho _\tau (y_{t2} - q_{t2}(\vartheta ))\) has its maximum value uniquely at an interior point \(\vartheta _0^*\) in \(\varTheta ^*\). \(\square\)
About this article
Cite this article
Lee, S., Kim, C.K. Test for conditional quantile change in general conditional heteroscedastic time series models. Ann Inst Stat Math 76, 333–359 (2024). https://doi.org/10.1007/s10463-023-00889-z
Received:
Revised:
Accepted:
Published:
Issue Date:
DOI: https://doi.org/10.1007/s10463-023-00889-z