Kyushu University Academic Staff Educational and Research Activities Database
List of Presentations
Hiroki Masuda Last modified date:2021.11.01

Professor / Department of Mathematical Sciences / Faculty of Mathematics


Presentations
1. Hiroki Masuda, On mixed-rates structure in Gaussian quasi-likelihood inference for Lévy SDE, Workshop on Statistical modeling for stochastic processes and related fields, 2021.09, [URL].
2. Hiroki Masuda, Estimation and selection for ergodic Lévy driven models, PVSeminar, 2021.09, [URL].
3. Hiroki Masuda, Lévy-Ornstein-Uhlenbeck regression, Asia-Pacific Seminar in Probability and Statistics (APSPS), 2021.06, [URL].
4. Shoichi Eguchi, Hiroki Masuda, Gaussian quasi-information criterion for ergodic Lévy driven model, Virtual 63rd ISI World Statistics Congress, 2021.07.
5. Hiroki Masuda, Inference for a class of non-ergodic non-Gaussian regression, The LiU Seminar Series in Statistics and Mathematical Statistics, 2021.02, We consider statistical inference for a class of non-ergodic locally stable regression models with parametric trend and scale coefficients, when the process is observed at high frequency and the local stable (activity) index is unknown. A detailed asymptotics of the associated (conditional) stable quasi-likelihood estimator is given. In particular, we show that the asymptotic property of the estimator is affected in an essential way by a sort of nonlinearity of the scale coefficient.
Also shown is how we can conduct descriptive model selection in the non-ergodic setup..
6. Hiroki Masuda, Alexei Kulik, LAD estimation of locally stable SDE, Computational and Methodological Statistics, 2020.12, Our goal is to prove the asymptotic (mixed) normality of the least absolute deviation (LAD) type estimator of locally stable SDE observed at high frequency, where the target drift coefficient may be nonlinear in both state variable and parameter. The proof essentially relies on the recently developed general representation result about small-time stable approximation for general locally stable processes. The result is a far-reaching extension of the previous study Masuda (Electronic Journal of Statistics, 2010)..
7. Hiroki Masuda, Noise estimation for ergodic Levy driven SDE and YUIMA package, Séminaire AgroParisTech, 2020.01, Levy driven stochastic differential equation (SDE) is a flexible building block for modeling non-Gaussian high-frequency data observed in many application fields such as biology and ecology. It is, however, a common knowledge that a closed form of the likelihood function is rarely available except for quite special cases, making estimation of characteristics of the driving Levy noise difficult. In this talk, we begin with an overview of the related previous studies, and then propose a multistep estimation procedure based on the Euler residuals constructed from the Gaussian quasi-maximum likelihood estimator (GQMLE). Specifically, we first estimate the parametric coefficient by the GQMLE, next approximate “unit" time increments of the driving noise by partially summing up the Euler residuals; this strategy would be useful when the underlying Levy process has a tractable unit-time density while having very complicated Levy measure. We will present large-sample properties of the proposed estimator, followed by how to implement it to the YUIMA package in R (https://yuimaproject.com). This talk is based on the ongoing joint work with Yuma Uehara (The Institute of Statistical Mathematics) and Lorenzo Mercuri (University of Milan)..
8. Hiroki Masuda, Mercuri, Lorenzo, Yuma Uehara, Noise estimation for ergodic Lévy driven SDE in YUIMA package, CMStatistics 2019, 2019.12, To describe non-Gaussian activity in high frequency data obtained from financial, biological, and technological phenomenon, Levy driven stochastic differential equations serve as good candidates. Since the closed form of its genuine likelihood is generally not obtained, the estimation of its driving noise is often done by empirical moment fittings with respect to its Levy measure. However, the measure sometimes takes complex form, and thus intractable. For such a problem, we consider the approximation of unit time increments of the driving noise based on the Euler residual. By making use of this approximation, we can conduct parametric estimation methods of the driving noise with bias correction. We will present its theoretical properties and show some numerical experiments..
9. Hiroki Masuda, On Lévy driven models in YUIMA, The 3rd YUIMA conference, 2019.06.
10. Hiroki Masuda, Simulation of diffusion processes; Lévy processes: basics and simulation; Lévy driven SDE: basics and simulation; Quasi-likelihood estimation of Lévy driven SDE, YUIMA summer school, 2019.06.
11. Hiroki Masuda, Nonlinear locally stable regression, Dynstoch meeting 2019, 2019.06, We consider statistical inference for a class of non-ergodic locally stable regression models with parametric trend and scale coefficients, when the process is observed at high frequency and the local stable (activity) index is unknown. A detailed asymptotics of the associated (conditional) stable quasi-likelihood estimator is given. In particular, we show that the asymptotic property of the estimator is affected in an essential way by a sort of nonlinearity of the scale coefficient, resulting in a significant generalization of the previous finding in Brouste and Masuda (2018, SISP). We will also mention how to implement the proposed statistics into the YUIMA package (2014)..
12. Hiroki Masuda, Locally stable regression with unknown activity index, CMStatistics 2018, 2018.12, Typically, transition of large-scale dependent data, such as those sampled at ultra high-frequency, are highly non-Gaussian. One of natural ways of modeling such data would be to use continuous-time stochastic processes driven by a non-Gaussian pure-jump noise. The related existing literature is, however, still far from being well-developed. In this talk, we present tailor-made quasi-likelihood inference results that can efficiently handle such locally and highly non-Gaussian statistical models with the activity index of the driving noise process being unknown. The model setup includes not only Markovian stochastic differential equations but also a class of semimartingale regression models. Of primary interest are cases where estimation target includes not only the rapidly varying scale structure but also the slowly varying trend one..
13. Hiroki Masuda, Optimal stable regression, The 5th Institute of Mathematical Statistics Asia Pacific Rim Meeting (APRM), 2018.06.
14. Hiroki Masuda, On stable regression, dynstoch 2018, 2018.06.
15. Shoichi Eguchi, Hiroki Masuda, Sampling stepsize in practice: fine tuning and/or estimation, Computational Aspects of Simulation and Inference for Stochastic Processes and the YUIMA Project, 2018.03.
16. Alexandre Brouste, Hiroki Masuda, Efficient estimation of stable Lévy process, ASC2018, Asymptotic Statistics and Computations, 2018.02.
17. Shoichi Eguchi, Hiroki Masuda, Modeling time scale in high-frequency data, CMStatistics 2017, 2017.12.
18. Hiroki Masuda, Local limit theorem in non-Gaussian quasi-likelihood inference, Asymptotic Statistics of Stochastic Processes and Applications XI, 2017.07, We consider parameter estimation of the finite-dimensional parameter in the stochastic differential equation (SDE) model driven by a highly non-Gaussian noise. We will present handy sufficient conditions for the L1-local limit theorem with convergence rate, which is the key assumption for the asymptotic mixed normality. The sufficient conditions are given only in terms of the driving Levy measure and/or the characteristic exponent of the driving noise. Specific examples satisfying them include stable, exponentially tempered $¥beta$-stable, and generalized hyperbolic Levy processes..
19. Hiroki Masuda, Stable quasi-likelihood regression, EcoSta 2017, 2017.06.
20. Hiroki Masuda, Shoichi Eguchi, Yuma Uehara, Lévy SDE inference in Yuima package, Dynstoch meeting 2017, 2017.04.
21. Hiroki Masuda, Remarks on Gaussian quasi-likelihood inference for Lévy driven SDE, ASC2017, Asymptotic Statistics and Computations, 2017.01.
22. Hiorki Masuda, Shoichi Eguchi, Yuma Uehara, Stepwise estimation and assessment of Lévy driven SDE, CMStatistics 2016, 2016.12.
23. Hiroki Masuda, Locally stable regression without ergodicity and finite moments, Hokkaido International Symposium "Recent Developments of Statistical Theory in Statistical Science", 2016.10.
24. Hiroki Masuda, Yusuke Shimizu, On regularized estimation of ergodic diffusion process, Advances in Statistics for Random Processes, 2016.09, We are concerned here with how to deduce the uniform integrability of a regularized, possibly sparse type quasi-likelihood estimator of an ergodic diffusion process observed at high frequency. For this purpose, we will make use of the general machinery for proving polynomial type uniform tail-probability estimate of a scaled M-estimator, showing how it can carry over to cases of multiple and mixed-rates asymptotics where associated statistical random fields not only may be non-differentiable, but also may fail to be locally asymptotically quadratic..
25. Hiroki Masuda, On Asymptotics of multivariate non-Gaussian quasi-likelihood, World Congress in Probability and Statistics, 2016.07, We consider (semi-)parametric inference for a class of stochastic differential equation (SDE) driven by a locally stable Levy process, focusing on multivariate setting and some computational aspects. The process is supposed to be observed at high frequency over a fixed time domain. This setting naturally gives rise to a theoretically fascinating quasi-likelihood which brings about a novel unified estimation strategy for targeting a broad spectrum of driving Levy processes. The limit experiment is mixed normal with a clean-cut random information structure, based on which it is straightforward to make several conventional asymptotic statistical decisions. The infill-asymptotics adopted here makes the popular Gaussian quasi-likelihood useless, while instead enabling us not only to incorporate any exogenous and/or observable endogenous data into the trend and/or scale coefficients without essential difficulty, but also to sidestep most crucial assumptions on the long-term stability such as ergodicity and moment boundedness. The proposed quasi-likelihood estimator is asymptotically efficient in some special cases..
26. Hiroki Masuda, On Asymptotics of multivariate non-Gaussian quasi-likelihood, The 4th Institute of Mathematical Statistics Asia Pacific Rim Meeting, 2016.06.
27. Hiroki Masuda, Shoichi Eguchi, On Schwarz type model comparison, Dynstoch meeting 2016, 2016.06.
28. Hiroki Masuda, Lévy in quasi-likelihood estimation of SDE, Statistics for Stochastic Processes and Analysis of High Frequency Data V, 2016.03, We try to give a clear whole picture about the local stable approximation in estimating a L\'{e}vy driven SDE under infill asymptotics without ergodicity. Our finding here is that the completely analogous strategy as in the local Gauss approximation in estimating a diffusion does a good job, when the activity degree is equal to or greater than 1 (the Cauchy-like case). The proposed estimator is indeed asymptotically efficient in some instances..
29. Hiroki Masuda, Yuma Uehara, Stepwise estimation of ergodic Lévy driven SDE, ASC2016: Asymptotic Statistics and Computations, 2016.02.
30. Hiroki Masuda, Computational aspects of estimating Lévy driven models, The 9th IASC-ARS conference, 2015.12, We consider estimation problem concerning stochastic differential equations driven by a Levy process with jumps. The model is supposed to be observed at high-frequency, allowing us to incorporate a small-time approximation of the underlying likelihood. An overview of some existing theories based on the Gaussian and non-Gaussian quasi-likelihoods is presented, together with their computational aspects. Also to be demonstrated is how to implement the theory in the YUIMA package: an R framework for simulation and inference of stochastic differential equations..
31. Hiroki Masuda, Shoichi Eguchi, Approximate Bayesian model comparison of LAQ models, 60th World Statistical Congress, ISI2015, 2015.07, For model-specification purpose, we study asymptotic behavior of the marginal quasi-log likelihood associated with a family of locally asymptotically quadratic (LAQ) statistical experiments. Our result entails a far-reaching extension of applicable scope of the classical approximate Bayesian model comparison due to Schwarz, with frequentist-view theoretical foundation. In particular, the proposed statistics can deal with both ergodic and non-ergodic stochastic-process models, where the corresponding $M$-estimator is of multi-scaling type and the asymptotic quasi-information matrix is random. Focusing on the ergodic diffusion model, we also deduce the consistency of the multistage optimal-model selection where we may select an optimal sub-model structure step by step, so that computational cost can be much reduced. We illustrate the proposed method by the Gaussian quasi-likelihood for diffusion-type models in details, together with several numerical experiments..
32. Hiroki Masuda, Yuma Uehara, Explicit bias correction in functional estimation of Lévy driven ergodic SDE, Dynstoch meeting 2015, 2015.05, We consider high frequency samples from ergodic L¥'{e}vy driven stochastic differential equation (SDE) with drift coefficient $a(x,¥alpha)$ and scale coefficient $c(x,¥gamma)$ involving unknown parameters $¥alpha$ and $¥gamma$. We suppose that the L¥'{e}vy measure $¥nu_{0}$, has all order moments but is not fully specified. We will prove the joint asymptotic normality of some estimators of $¥alpha$, $¥gamma$ and a class of functional parameter $¥int¥varphi(z)¥nu_0(dz)$, which are constructed in a two-step manner: first, we use the Gaussian quasi-likelihood for estimation of $(¥al,¥gam)$; and then, for estimating $¥int¥varphi(z)¥nu_0(dz)$ we makes use of the method of moments based on the Euler-type residual with the the previously obtained quasi-likelihood estimator..
33. Hiroki Masuda, On variants of stable quasi-likelihood for Levy driven SDE, Statistique Asymptotique des Processus Stochastiques X, 2015.03.
34. Hiroki Masuda, Shoichi Eguchi, On quasi-BIC for general LAQ model, ERCIM 2014, 2014.12.
35. Hiroki Masuda, On sampling problem for pure-jump SDE , 3rd APRM, Taipei, 2014.07.
36. Hiroki Masuda, Yusuke Shimizu, Remark on the large deviation inequality in mixed-rates asymptotics, ASC2014 Asymptotic Statistics and Computations 2014, 2014.03, [URL], We will show how the polynomial type large deviation inequality (PLDI) of Yoshida (2011) can carry over to the ``mixed-rates'' M-estimation where the statistical random field in question may have components converging at different rates; see Radchenko (2008) for a general framework. It will be shown that the PLDI criterion of Yoshida (2011) can apply directly in ``moderate'' mixed-rates cases while it requires some modification for cases of ``drastically-different'' rates, where the key partially locally asymptotically quadratic (PLAQ) structure no longer holds true. For example, many of the sparse-type estimation procedures may fall into this type of asymptotics. Our conditions may not be minimal, but it does not seem to be quite straightforward to soften them. Although our claims do not offer methodologically new, it does provide a theoretically deeper understanding on the recently highlighted sparse estimation. .
37. Hiroki Masuda, LAD-based estimation of locally stable Ornstein-Uhlenbeck processes, Waseda International Symposium on "Stable Process, Semimartingale, Finance & Pension Mathematics", 2014.03, [URL], The LAD type estimator for discretely observed Levy driven OU process is much more efficient than the LSE type one. We prove that the proposed estimator under a random norming is asymptotically standard-normally distributed, making construction of confidence intervals easy..
38. Hiroki Masuda, Stable quasi-likelihood: Methodology and computational aspects, ERCIM 2013 London, 2013.12, [URL], We consider the semi-parametric model described by the parametric locally stable pure-jump stochastic differential equation. We wish to estimate the parametric coefficients based on a high-frequency sample over a fixed interval. In this talk, we introduce a novel, tailor-made estimator based on the stable approximation of the one-step transition distribution. Under suitable regularity conditions, it is shown that the proposed estimator is asymptotically mixed-normal. The result reveals that, in case of the stable-like driving Levy process, the proposed estimator is much more efficient than the conventional Gaussian quasi-maximum likelihood estimator, which requires the large-time asymptotics and leads to a slower rates of convergence. Nevertheless, evaluation of the proposed estimator is computationally more involved compared with the Gaussian case. Also discussed in some detail is the computational aspects of the proposed methodology.
.
39. Hiroki Masuda, Multi-step estimation procedure for stable Ornstein-Uhlenbeck processes, Stochastic processes and their statistics in Finance in Okinawa, 2013.10, [URL], We consider estimation of a non-Gaussian stable Ornstein-Uhlenbeck process based on discrete-time infill sampling. We show that the asymptotic mixed normality holds true for the associated statistical experiments, and discuss how to construct a practical estimator in an easy way without much information loss. As for estimation of all the unknown parameters involved, we will propose a multistep estimation procedure, expected to result in a somewhat more stable performance compared with simultaneous optimization..
40. Osaka University, [URL].
41. Hiroki Masuda, On statistical inference for Levy-driven models, The 59th World Statistics Congress (WSC), 2013.08, [URL], 保険数理分野では局所安定型確率微分方程式によるモデリングが有用である.モデルを適合させる対象期間を固定しつつ統計的分布論の理論基盤を確保できるという点において,ノイズの非正規性が如実に現れる当該分野での推測問題に新たな視点・展開を与えた..
42. Hiroki Masuda, Estimation of stable-like stochastic differential equations, 29th European Meeting of Statisticians, 2013.07, [URL], We consider the stochastic differential equation of pure-jumps type with parametric coefficients. We wish to estimate the unknown parameters based on a discrete-time but high-frequency sample. A naive way would be to use the Gaussian quasi likelihood. However, although the Gaussian quasi likelihood is known to be well-suited for the case of diffusions, it leads to asymptotically suboptimal estimator in the pure-jump case; in particular, the Gaussian quasi-maximum likelihood estimation inevitably needs a large-time asymptotics. In this talk, we will introduce another kind of quasi-maximum likelihood estimator based on the local-stable approximation of the one-step transition distribution; the proposed estimation procedure is a pure-jump counterpart to the Gaussian quasi-maximum likelihood estimation. Under some regularity conditions, we will show the asymptotic mixed normality of the proposed estimator, revealing that the proposed estimator is asymptotically much more efficient than the Gaussian quasi-maximum likelihood estimator..
43. Hiroki Masuda, On optimal estimation of stable Ornstein-Uhlenbeck processes, Dynstoch meeting 2013, 2013.04, [URL], Ornstein-Uhlenbeck (OU) processes driven by a Levy process form a particular tractable class of Markovian stochastic differential equations with jumps. Among them, the non-Gaussian stable driven ones, the study of which dates back to Doob's work in 1942, are known to have a pretty inherent character. Especially, a special property of stable integrals allows us to exactly generate the discrete-time sample from the process, and more importantly, to study in a transparent way the likelihood ratio associated with discrete-time sampling. We are concerned with optimal estimation of the stable OU processes observed at high-frequency. We clarify that, due to the infinite-variance character of the model, the likelihood ratio exhibits entirely different asymptotic behaviors according to whether or not the terminal sampling time tends to infinity. When the terminal time is a fixed time, we present the LAMN (Local Asymptotic Mixed Normality) structure of the statistical model, entailing the notion of asymptotic efficiency of a regular estimator. Also presented is how to construct some simple rate-efficient estimators having asymptotic mixed normality, together with numerical experiments..
44. Hiroki Masuda, Non-Gaussian quasi-likelihoods for estimating jump SDE, 8th World Congress in Probability and Statistics, 2012.07, We consider a stochastic differential equation driven by a stable-like Levy process, which is observed at high frequency.
In this talk, we will introduce a quasi-maximum likelihood estimator based on the local-stable approximation of the transition laws.
This is a pure-jump counterpart to the local-Gauss contrast function, well-suited for the case of diffusions.
Under some regularity conditions, we will present asymptotic distribution results, which is entirely different from the Gaussian quasi-likelihood case and much more efficient.
In particular, the rate of convergence of the estimator obtained is much better
and they are jointly asymptotically normal and mixed-normal according as the terminal sampling tends to infinity or not. .
45. Hiroki Masuda, Non-Gaussian quasi likelihood in estimating jump SDE, 2nd Asian Pacific Rim Meeting, 2012.07, 非正規安定レヴィ過程で微小時間近似できる確率微分方程式モデルの推定問題を考察した.当該モデルでは従来の正規型擬似最尤推定は効率が悪いことが知られており,新たな推定手法が要求される.筆者は,データ増分の非正規安定近似を介した新しい擬似尤度推定法を考案し,その漸近挙動を導出した.特に,ドリフト推定量の有界時間区間上での漸近混合正規性,および推定量の収束率の改善など,正規型では決して得られない(好ましい)現象が明らかとなった..
46. Hiroki Masuda, Local-stable contrast function, Dynstoch meeting 2012, 2012.06, We consider a stochastic differential equation driven by a stable-like Levy process, which is observed at high frequency.
In this talk, we will introduce a quasi-maximum likelihood estimator based on the local-stable approximation of the transition laws.
This is a pure-jump counterpart to the local-Gauss contrast function, well-suited for the case of diffusions.
Under some regularity conditions, we will present asymptotic distribution results, which is entirely different from the Gaussian quasi-likelihood case and much more efficient.
In particular, the rate of convergence of the estimator obtained is much better
and they are jointly asymptotically normal and mixed-normal according as the terminal sampling tends to infinity or not.
.