SEARCH
Search Details
SUEISHI NaoyaGraduate School of Economics / Division of EconomicsProfessor
Research activity information
■ Award■ Paper
- Informa UK Limited, Jan. 2025, Journal of Business & Economic Statistics, 1 - 12[Refereed]Scientific journal
- Aug. 2024, Econometric Theory, English[Refereed]Scientific journal
- Feb. 2023, 国民経済雑誌, 227(2) (2), 15 - 27変数選択後の統計的推測
- Springer Science and Business Media LLC, Apr. 2022, The Japanese Economic Review, 73(2) (2), 299 - 324[Refereed]Scientific journal
- Jun. 2021, 国民経済雑誌, 223(6) (6), 73 - 88条件付きモーメント制約モデルの効率性限界
- Oct. 2019, Kokumin-Keizai Zasshi, 220(4) (4), JapaneseRegularization Parameter Selection Methods for LassoResearch institution
- May 2019, Economics Letters, 178, 1 - 4, English[Refereed]Scientific journal
- Mar. 2019, Econometrics, 7(1) (1), English[Refereed]Scientific journal
- Public Library of Science, May 2018, PLoS ONE, 13(5) (5), English[Refereed]Scientific journal
- Dec. 2017, 国民経済雑誌, Japaneseモーメント制約モデルの効率性限界と経験尤度推定量の漸近理論に関する考察Research institution
- Oct. 2017, ECONOMETRIC THEORY, 33(5) (5), 1242 - 1258, English[Refereed]Scientific journal
- Sep. 2017, JAPANESE ECONOMIC REVIEW, 68(3) (3), 352 - 363, English[Refereed]Scientific journal
- Jan. 2016, ECONOMICS LETTERS, 138, 57 - 59, English[Refereed]Scientific journal
- Jul. 2013, Econometrics, 1(2) (2), 141 - 156, EnglishGeneralized Empirical Likelihood-Based Focused Information Criterion and Model Averaging[Refereed][Invited]Scientific journal
- Mar. 2013, Economics Letters, 118(3) (3), 509 - 511, English[Refereed]Scientific journal
- This paper proposes efficient estimation methods of unknown parameters when frequencies as well as local moments are available in grouped data. Assuming the original data is an i.i.d. sample from a parametric density with unknown parameters, we obtain the joint density of frequencies and local moments, and propose a maximum likelihood (ML) estimator. We further compare it with the generalized method of moments (GMM) estimator and prove these two estimators are asymptotically equivalent in the first order. Based on the ML method, we propose to use the Akaike information criterion (AIC) for model selection. Monte Carlo experiments show that the estimators perform remarkably well, and AIC selects the right model with high frequency.THE JAPAN STATISTICAL SOCIETY, Mar. 2008, Journal of the Japan Statistical Society, 38(1) (1), 131 - 143, English[Refereed][Invited]Scientific journal
- Long Memory Process and Heavy-Tailed DistributionThis paper surveys the models and estimation methods for the long memory processes with heavy-tailed distribution. There is a growing interest in these models in various fields of study such as network traffic and finance. We firstly introduce the standard Gaussian long memory processes: fractional Gaussian noise model and Gaussian FARIMA model. We extend them to the stable processes: linear fractional stable noise model and stable FARIMA model. A stable distribution is a generalization of the central limit theorem and is characterized as a limit distribution of normalized sum of independently and identically distributed random variables. Variances do not need to be finite in the stable distribution and is useful for modeling phenomena caused by heavy-tailed distributions. Since these models have a stable marginal distribution, they can capture not only the long memory but also the heaviness of tails which cannot be described by Gaussian processes.日本統計学会, Mar. 2006, 日本統計学会誌, 35(2) (2), 213 - 232, Japanese[Refereed]Scientific journal
- 2005, MODSIM 2005: INTERNATIONAL CONGRESS ON MODELLING AND SIMULATION: ADVANCES AND APPLICATIONS FOR MANAGEMENT AND DECISION MAKING, 926 - 932, EnglishSemiparametric Estimators for Conditional Moment Restrictions Containing Nonparametric Functions: Comparison of GMM and Empirical Likelihood Procedures[Refereed]International conference proceedings
- 2005, MODSIM 2005: INTERNATIONAL CONGRESS ON MODELLING AND SIMULATION: ADVANCES AND APPLICATIONS FOR MANAGEMENT AND DECISION MAKING, 953 - 959, EnglishEstimation of Levy Processes in Mathematical Finance: A Comparative Study[Refereed]International conference proceedings
- Nov. 2022, arXiv
- Dec. 2017, SSRN, EnglishOn the Convergence Rate of the SCAD-Penalized Empirical Likelihood EstimatorTechnical report
- Apr. 2015, 日本労働研究雑誌, Japaneseサンプルセレクションとセルフセレクション[Invited]Introduction scientific journal
- 日本評論社, Oct. 2013, 経済セミナー = The keizai seminar, (674) (674), 76 - 85, Japanese中級計量経済学 : 現代的手法へのいざない(第4回)分位点回帰
- 日本評論社, Apr. 2013, 経済セミナー = The keizai seminar, (671) (671), 71 - 81, Japanese中級計量経済学 : 現代的手法へのいざない(第1回)線形回帰とOLS
- EFFICIENT ESTIMATION AND MODEL SELECTION FOR GROUPED DATA WITH LOCAL MOMENTS(
CELEBRATION VOLUME FOR AKAIKE) This paper proposes efficient estimation methods of unknown parameters when frequencies as well as local moments are available in grouped data. Assuming the original data is an i.i.d. sample from a parametric density with unknown parameters, we obtain the joint density of frequencies and local moments, and propose a maximum likelihood (ML) estimator. We further compare it with the generalized method of moments (GMM) estimator and prove these two estimators are asymptotically equivalent in the first order. Based on the ML method, we propose to use the Akaike information criterion (AIC) for model selection. Monte Carlo experiments show that the estimators perform remarkably well, and AIC selects the right model with high frequency.Japan Statistical Society, 01 Jun. 2008, Journal of the Japan Statistical Society, 38(1) (1), 131 - 143, English - Efficiency Improvement by Local Moments in Grouped Data AnalysisThis paper proposes an efficient density estimation method for analyzing grouped data when local moments are given. We use the generalized method of moments (GMM) estimator of Hansen (1982) to incorporate the information contained in the local moments. We show that our estimator is more efficient than the classical maximum likelihood estimator for grouped data. We also construct a specification test statistic based on moment conditions. Monte Carlo experiments suggest that our estimator performs remarkably well and the specification test has good size properties even in finite samples.Kyoto University, Jun. 2006, 21COE Interfaces for Advanced Economic Analysis Discussion Paper, 107, 1 - 12, English
- 日本評論社, Apr. 2024, Japanese, ISBN: 9784535540484データ駆動型回帰分析 : 計量経済学と機械学習の融合
- Single work, 日本評論社, Jul. 2015, Japanese計量経済学 ミクロデータ分析へのいざないTextbook
- 日本経済学会春季大会, May 2023A Misuse of Specification Tests
- 日本経済学会春季大会, May 2020Large Sample Justifications for the Bayesian Empirical LikelihoodOral presentation
- 関西計量経済学研究会, Jan. 2020Large Sample Justifications for the Bayesian Empirical Likelihood
- The 2nd Annual International Conference on Applied Econometrics in Hawaii, Jun. 2016, English, International conferencePenalized empirical likelihood estimation and model selection for high-dimensional misspecified moment restriction models[Invited]Nominated symposium
- 日本学術振興会, 科学研究費助成事業, 基盤研究(C), 神戸大学, 01 Apr. 2024 - 31 Mar. 2027意思決定のための因果推論の理論
- Japan Society for the Promotion of Science, Grants-in-Aid for Scientific Research, Grant-in-Aid for Scientific Research (B), Hosei University, 01 Apr. 2022 - 31 Mar. 2025Machine Learning methods for Econometric analysis機械学習の発想や手法を計量経済学に応用する研究を遂行した。まず、機械学習手法に由来するParameter Tyingのアイディアを計量時系列分析に応用することでtying maximum likelihood estimation (TMLE)の方法を開発した。この研究 は、複数の時系列変量の観測期間の長さが異なり、極端に観測期間の短い変量がある場合に、推定パフォーマンスを高めることを目的とする。観測期間の長い時系列と短い時系列のパラメータにある程度の共通性があるならば、TMLEは、短い時系列のパラメータと長い時系列のパラメータをTying (結びつける)ことにより、長い時系列のデータに含まれる短い時系列のパラメータを推定するために有用な情報を移転でき、推定の精度を改善する。本研究ではTMLEの漸近的性質と有限標本の性質を導出し、Tyingの強さを調整するためのTuning Parameterの選択法を開発した。Machine Collaborationに関する研究では計算アルゴリズムのスケーラビリティの問題に直面しており、その解決法を検討している。またMachine Collaborationの分類問題への応用を検討している。さらに、機械学習のKmeans アルゴリズムを利用してグループ分けを行う線形パネルデータモデルの研究と、double machine learning proceduresによるパネルデータモデルの推定法の研究を行なった。もう一つの研究として、高次元データを用いたLassoと呼ばれる統計的推測の方法のバイアスを修正するためのdebiased Lassoと呼ばれる方法の新しいチューニングパラメータの選択方法を提案し、一定の条件の下で、分散を発散させることなく既存手法よりもバイアスを小さくすることが可能であることを示した。
- Japan Society for the Promotion of Science, Grants-in-Aid for Scientific Research, Grant-in-Aid for Scientific Research (C), Kobe University, 01 Apr. 2021 - 31 Mar. 2024Efficiency bound for moment restriction models with time series observations本研究では、モーメント制約によって定式化される統計モデル(モーメント制約モデル) の効率性限界の導出を目指す。一般に、統計モデルに含まれる有限次元パラメータの推定に おいては、いかなる正則な推定量を用いても下回ることのできない漸近分散の下限が存在し、その下限は効率性限界と呼ばれる。独立で同一の分布に従う(i.i.d.)データの場合には、モーメント制約モデルの効率性限界は既知であるが、時系列データの場合の効率性限界は未知であり、計量経済学の理論研究における未解決問題として残されている。本研究では、時系列モーメント制約モデルの効率性限界を導出する。 効率性限界を導出するためには、モデルの局所漸近正規性(LAN: local asymptotic normality)と呼ばれる性質を示すことが鍵となる。本年度は、時系列モーメント制約モデルのLAN性を示す前段階として、i.i.d.のケースでモーメント制約モデルと条件付きモーメント制約モデルのLAN性を示す方法について考察した。 また、派生研究として、モーメント制約によって定式化されるモデルの特定化検定と、モデルに含まれるパラメータの漸近的に効率的な推定量の関係性について考察した。局所的な定式化の誤りの下での検定統計量と推定量の漸近的な性質を調べた結果、検定統計量を用いて検出できる分布のずれの方向と漸近的に効率的な推定量にバイアスをもたらすような分布のずれの方向が直交することが示された。これは特定化検定を用いても、推定量にバイアスが生じるかどうかを判断するのは不可能であることを示すものである。
- Japan Society for the Promotion of Science, Grants-in-Aid for Scientific Research, Grant-in-Aid for Scientific Research (B), Kyoto University, 01 Apr. 2019 - 31 Mar. 2023Microeconometric analysis for causal inference and its applications to EBPMWe developed a causal inference method for situations where there is heterogeneity in intervention effects across unknown groups, as well as non-compliance with intervention allocation and network spillovers. We clarified the minimax theory of nonparametric specification tests of the functional form of IV models. We developed an estimation method with low computational burden and asymptotic efficiency where the likelihood function is unknown but the characteristic function is known. The relationship between station spacing and land use along public transport systems was investigated, and a counterfactual analysis was conducted on the effect of changes in spacing on urban compactness. Using big data from POS, the study showed that the seasonality of infection prevention products disappeared with the COVID-19 disaster and that there was a substantial decline in sales of outing-related items, while sales of unrelated items were more affected by the shock of the consumption tax rise.
- Japan Society for the Promotion of Science, Grants-in-Aid for Scientific Research, Grant-in-Aid for Scientific Research (C), Kobe University, Apr. 2018 - Mar. 2021, Principal investigatorThis study investigated the asymptotic properties of the Bayesian empirical likelihood (BEL), which uses the empirical likelihood as an alternative to a parametric likelihood for Bayesian inference. There are two main findings. First, the limiting posterior distribution of the BEL is the same as that of a parametric Bayesian method that uses the likelihood of a least favorable model of the moment restriction model. Second, the limiting posterior distribution is also the same as that of a semiparametric Bayesian method.Competitive research funding
- Japan Society for the Promotion of Science, Grants-in-Aid for Scientific Research, Grant-in-Aid for Scientific Research (C), Kobe University, Apr. 2015 - Mar. 2018, Principal investigatorPenalized estimation is a useful technique for variable section when the number of potential variables is large. A crucial issue in penalized estimation is the selection of the regularization parameter. However, there has been little study on the selection method of the regularization parameter for the penalized estimation of moment restriction models. This study proposes a new information criterion, which we call the empirical likelihood information criterion, to select the regularization parameter of the penalized empirical likelihood estimator. On the basis of the idea of the Akaike information criterion, our information criterion is derived as an asymptotically unbiased estimator for the Kullback-Leibler information criterion.Competitive research funding
- Japan Society for the Promotion of Science, Grants-in-Aid for Scientific Research, Grant-in-Aid for Young Scientists (B), 01 Apr. 2013 - 31 Mar. 2016Bayesian estimation of moment inequality models using empirical likelihoodThis study proposed a Bayesian estimation method that uses an empirical likelihood as a likelihood function, and applied the method to the estimation of moment inequality models. I showed asymptotic properties of the posterior density. Furthermore, I investigated finite sample validity of using an empirical likelihood as a likelihood of Bayes estimation. Moreover, this study derived a least favorable submodel of conditional moment restriction models. The result suggests the efficiency bound and an asymptotically efficient estimator. The resulting estimator can be viewed as an empirical likelihood estimator for conditional moment restriction models.
- Japan Society for the Promotion of Science, Grants-in-Aid for Scientific Research, Grant-in-Aid for Young Scientists (B), Kyoto University, 2011 - 2012Focused information criterion for semiparametric modelsThe goal of model selection is to select a “best” model in a data-driven way. However, the best model generally differs for different intended use of the model. This study develops an empirical likelihood-based model selection method for moment restriction models that is designed to obtain a good estimate for a specific parameter of interest. This study also investigates a model averaging method for moment restriction models that minimizes the mean squared error of the estimator.
- 日本学術振興会, 科学研究費助成事業, 基盤研究(C), 椙山女学園大学, 2011 - 2012金融時系列分析におけるノンパラメトリック・ボラティリティ推定ボラティリティとは収益率の分散(あるいま標準偏差)の事を言うが,金融データの時系列分析では収益率の系列を調べても,通常,何ら特性が見つからない事が一般に知られている.特に,効率的な市場だと,収益率はランダム・ウォークに従うとされる.この研究では金融時系列におけるボラティリティの分析法の研究を行う.特にティック・データが利用できる状況において,効率的かつ実際的なボラティリティの推宿法を求めてきた.ティック・データを用いたボラティリティ分析でよノイズの処理が理論的には大きな関心を持たれているが,時系列処理で,ノイズをフィルターする手法を考案してきた. ボラティリティは自在に変化し,ノイズは独立同一分布を持つという通常の仮定の下ではボラティリティは観測個数にのみ依存する値となる.この1生質を使って,間引き標本などの手法を用いてノイズを除去しようとする試みは行われてきたが,本研究では二次モーメントを基礎とするノイズ除去法を追求した. 理論的な研究を進めてきたが,数系列を用いたシミュレーションも行った.しかし,このようなシミュレーションは常に理論的な結果をサポートする事になるので,実際の金融データを使った実証的な検証も行った.この手順は非常に重要で,理論的な分析がもたらすある種のモデル・スペシフィケーション・バイアスを除去することができた幅の広い誤差分布においても従来の分析法が維持できるか否かを検討したが,明確な結果は得ることができなかった.代表的な系列に限定してもデータ購入の費用が高額であった.データについてはこれからも利用していく.