Workshops

The Applied Statistics Workshop 2019

※統計数理研究所リスク解析戦略研究センター金融・保険リスク研究プログラムとの共催ワークショップ

※ 2019年10月9日現在の予定です。

日時

2019年10月11日(金 Friday) 16:50-18:35

場所

東京大学大学院経済学研究科 学術交流棟 (小島ホール)1階 第1セミナー室[地図]
in Seminar Room 1 on the 1st floor of the Economics Research Annex (Kojima Hall) [Map]

報告

矢田和善(筑波大学)

「高次元の統計学:高次元PCAとその応用」

要旨(Abstract) 従来型のPCA手法は高次元のもと不一致性をもつことが知られている。それに対し、高次元データにおいても一致性を有するような高次元PCA法が、Aoshima and Yataの一連の研究によって提案された。本講演では、まず、高次元PCA法について解説する。さらに、高次元PCAと高次元統計解析法を融合させた新たな高次元統計解析の展開を紹介する。具体的には、高次元PCAを用いた二標本検定や判別分析について実データを交えながら解説する。本研究は青嶋教授(筑波大学)との共同研究です。  

日時

2019年10月18日(金 Friday) 16:50-18:35

場所

東京大学大学院経済学研究科 学術交流棟 (小島ホール)1階 第1セミナー室[地図]
in Seminar Room 1 on the 1st floor of the Economics Research Annex (Kojima Hall) [Map]

報告

間野修平(統計数理研究所)

"Direct Sampler from Toric Models with Computational Algebra"

要旨(Abstract) For a toric model (hierarchical log-linear model) we usually consider sampling with MCMC, but it is not always easy to construct irreducible Markov chain. Diaconis and Sturmfels (1998, Ann. Stat.) showed that a Groebner basis of a toric ideal of the polynomial ring provides a basis of moves, and such bases are called Markov bases. However, by considering holonomic ideal of the ring of differential operators, the speaker showed that a direct and i.i.d. sampling from a toric model is possible and provided the algorithm (2017, Electron. J. Stat.). In this talk, I will introduce some of toric models which have been considered to hamper direct sampling. The examples include non-null models, non-decomposable graphical models, non-graphical models, and Young diagrams of non-exchangeable partitions. Then, I will introduce the direct sampling algorithm and explain how it works for such models. Finally, I will explain mathematical challenges emerging in replacing an MCMC sampler by the direct sampler. This talk is based on collaboration with Professor Nobuki Takayama at Kobe University.  

日時

2019年11月1日(金 Friday) 16:50-18:35

場所

東京大学大学院経済学研究科 学術交流棟 (小島ホール)1階 第1セミナー室[地図]
in Seminar Room 1 on the 1st floor of the Economics Research Annex (Kojima Hall) [Map]

報告

Ilia Negri (Bergamo University, Italy)

TBA

要旨(Abstract)  

日時

2019年11月22日(金 Friday) 16:50-18:35

場所

東京大学大学院経済学研究科 学術交流棟 (小島ホール)1階 第1セミナー室[地図]
in Seminar Room 1 on the 1st floor of the Economics Research Annex (Kojima Hall) [Map]

報告

村上大輔(統計数理研究所)

TBA

要旨(Abstract)  

日時

2019年12月6日(金 Friday) 16:50-18:35

場所

東京大学大学院経済学研究科 学術交流棟 (小島ホール)1階 第1セミナー室[地図]
in Seminar Room 1 on the 1st floor of the Economics Research Annex (Kojima Hall) [Map]

報告

星野伸明(金沢大学)

TBA

要旨(Abstract)  

日時

2019年12月20日(金 Friday) 16:50-18:35

場所

東京大学大学院経済学研究科 学術交流棟 (小島ホール)1階 第1セミナー室[地図]
in Seminar Room 1 on the 1st floor of the Economics Research Annex (Kojima Hall) [Map]

報告

野間久史(統計数理研究所)

TBA

要旨(Abstract)  

日時

2019年1月10日(金 Friday) 16:50-18:35

場所

東京大学大学院経済学研究科 学術交流棟 (小島ホール)1階 第1セミナー室[地図]
in Seminar Room 1 on the 1st floor of the Economics Research Annex (Kojima Hall) [Map]

報告

Chris Glynn (University of New Hampshire)

TBA

要旨(Abstract)  

<以下本年度終了分>

 

日時

2019年4月19日(金 Friday) 16:50-18:35

場所

東京大学大学院経済学研究科 学術交流棟 (小島ホール)1階 第1セミナー室 [地図]
in Seminar Room 1 on the 1st floor of the Economics Research Annex (Kojima Hall) [Map]

報告

福元健太郎(学習院大学)

"Non-Ignorable Attrition in Pairwise Randomized Experiments"

要旨(Abstract)  The pairwise randomized experiments enable robust and efficient causal inference. What if outcomes of some units are missing? One way is to delete missing units and calculate difference-in-means (unitwise deletion estimator, UDE). Another method is to delete the other unit in the pair as well (pairwise deletion estimator, PDE). UDE is biased. Some scholars argue that PDE is unbiased, while opponents criticize that PDE is also biased if attrition is non-ignorable and PDE is less efficient than UDE. By using the potential outcome framework, this study formally shows that PDE can be biased but more efficient than UDE; the pairwise variance estimator of PDE is unbiased in the super-population. I argue that it is easier to interpret PDE as a causal effect than UDE. I also propose a new variance estimator. Finally, in order to show how PDE and UDE work, an application is demonstrated. This paper recommends PDE.

日時

2019年4月26日(金 Friday) 16:50-18:35

場所

東京大学大学院経済学研究科 学術交流棟 (小島ホール)1階 第1セミナー室 [地図]
in Seminar Room 1 on the 1st floor of the Economics Research Annex (Kojima Hall) [Map]

報告

Mike So (Hong Kong University of Science and Technology)

"Efficient Estimation of High-Dimensional Dynamic Covariance by Risk Factor Mapping: Applications for Financial Risk Management"

要旨(Abstract)  This paper aims to explore a modified method of high-dimensional dynamic variance-covariance matrix estimation via risk factor mapping, which can yield a dependence estimation of asset returns within a large portfolio with high computational efficiency. The essence of our methodology is to express the time-varying dependence of high-dimensional return variables using the co-movement concept of returns with respect to risk factors. A novelty of the proposed methodology is to allow mapping matrices, which govern the co-movement of returns, to be time-varying. We also consider the exible modeling of risk factors by a copula multivariate generalized autoregressive conditional heteroscedasticity (MGARCH) model. Through the proposed risk factor mapping model, the number of parameters and the time complexity are functions of a small number of risk factors instead of the number of stocks in the portfolio, making our proposed methodology highly scalable. We adopt Bayesian methods to estimate unknown parameters and various risk measures in the proposed model. The proposed risk mapping method and financial applications are demonstrated by an empirical study of the Hong Kong stock market. The assessment of the effectiveness of the mapping via risk measure estimation is also discussed.

日時

2019年5月24日(金 Friday) 16:50-18:35

場所

東京大学大学院経済学研究科 学術交流棟 (小島ホール)1階 第1セミナー室 [地図]
in Seminar Room 1 on the 1st floor of the Economics Research Annex (Kojima Hall) [Map]

報告

星野崇宏(慶応大学)

「潜在的結果変数の同時分布と因果効果の異質性の識別および推定について」

要旨(Abstract) ルービンの因果効果およびその異質性を推定することは、政策評価や最適割当な ど様々な分野で重要な課題である。但し潜在的結果変数の差のquantileなど、一 般に重要な量を推定するためには潜在的結果変数の同時分布を識別することが必 要となる。しかし無作為化実験であってもこの同時分布を識別することはできな い。 本研究では同時分布を識別できる条件を示す。さらにこれがノンコンプライアン スのある無作為化実験や外部情報が利用できる状況などで成立することを示し、 その推定方法を提案する。シミュレーションデータを用いて推定の精度を議論し、 職業訓練プログラムデータに応用した解析例を示す。 本研究は高畑圭祐氏(慶應義塾大学大学院)との共同研究である。 

日時

2019年6月7日(金 Friday) 16:50-18:35

場所

東京大学大学院経済学研究科 学術交流棟 (小島ホール)1階 第1セミナー室 [地図]
in Seminar Room 1 on the 1st floor of the Economics Research Annex (Kojima Hall) [Map]

報告

北村祐一 (Yale University)

"Methods for nonparametric counterfactual analysis with (or without) convex structure"

要旨(Abstract)  Abstract: Providing counterfactuals given an economic model, while avoiding ad hoc assumptions is an important task. In this talk I argue that there are many models that satisfy certain convexity properties, which, in turn, enable us to calculate counterfactuals and carry out inference about them under minimal assumptions. This typically yield predictions in terms of bounds, and we need inferential procedures that address non-standard features associated with them, but at the same time practically implementable. These tasks typically require optimization in very high dimensions, but they can be carried out under the convexity properties, combined with appropriate algorithms and recent advances in computational resources. Various examples will be discussed. This talk will be based on joint projects with Rahul Deb, Lucas de Lima, Myrto Kalouptsidi, Eduardo Souza Rodrigues, Joerg Stoye and John Quah.

日時

2019年7月5日(金 Friday) 16:50-18:35

場所

東京大学大学院経済学研究科 学術交流棟 (小島ホール)1階 第1セミナー室 [地図]
in Seminar Room 1 on the 1st floor of the Economics Research Annex (Kojima Hall) [Map]

報告

國濱剛(関西学院大学)

"Bayesian factor models for probabilistic cause of death assessment with verbal autopsies"

要旨(Abstract)  The distribution of deaths by cause provides crucial information for public health planning, response, and evaluation. About 60% of deaths globally are not registered or given a cause, limiting our ability to understand disease epidemiology. Verbal autopsy (VA) surveys are increasingly used in such settings to collect information on the signs, symptoms, and medical history of people who have recently died. This article develops a novel Bayesian method for estimation of population distributions of deaths by cause using verbal autopsy data. The proposed approach is based on a multivariate probit model where associations among items in questionnaires are flexibly induced by latent factors. Using the Population Health Metrics Research Consortium labeled data that include both VA and medically certified causes of death, we assess performance of the proposed method. Further, we estimate important questionnaire items that are highly associated with causes of death. This framework provides insights that will simplify future data collection.

日時

2019年7月19日(金 Friday) 16:50-18:35

場所

東京大学大学院経済学研究科 学術交流棟 (小島ホール)1階 第1セミナー室 [地図]
in Seminar Room 1 on the 1st floor of the Economics Research Annex (Kojima Hall) [Map]

報告

城田慎一郎(理化学研究所)

Robust Bayesian Nonparametric Inference for Heterogeneous Treatment Effects

要旨(Abstract)  近年、医療・経営分野において、異質因果効果の推定が注目を集めており、様々な機械学習・統計手法が提案されている。本発表では、異質因果効果の推定手法として、ベイズノンパラメトリック手法、特にガウス過程を用いた推定手法を紹介する。動機づけとなるデータとして、特にレセプトデータに代表される大規模医療データへの応用を念頭におき、サンプルセットの持つ異質性を考慮した手法を提案し、シミュレーション実験による検証を行う。

日時

2019年9月4日(水 Wednesday) 17:00-18:30  ※日時に注意

場所

東京大学大学院経済学研究科 学術交流棟 (小島ホール)1階 第1セミナー室 [地図]
in Seminar Room 1 on the 1st floor of the Economics Research Annex (Kojima Hall) [Map]

報告

Francis Hui (Australian National University)
"GEE-assisted variable selection for latent variable models: making the most of zero consistency"

要旨(Abstract)  In many disciplines, it is becoming common to collect and analyze multivariate or multi-response data. For example, this project is motivated by the Southern Ocean Continuous Plankton Recorder (SO-CPR) survey, an annual survey which collects presence-absence data on zooplankton assemblages in the Southern Ocean. One of the primary goals of the SO-CPR survey is to identify important environmental factors driving the species communities’ distribution, while accounting for biotic affects such as species interactions. An increasingly popular approach for analyzing multivariate data in ecology, among other disciplines, is generalized linear latent variable models (GLLVMs), which utilizes latent variables to parsimoniously account for residual between species correlations. However, estimation let alone variable selection for GLLVMs presents a major computational challenge, since the marginal likelihood function does not possess a closed form. To overcome this problem, we propose utilizing marginal generalized estimation equations (GEEs) to speed up inference on GLLVMs. Focusing on multivariate binary data, we show that GEEs are zero consistent for GLLVMs i.e., all truly zero (non-zero) conditional coefficients in a GLLVM will be consistently estimated as zero (non-zero) by the marginal coefficients in a GEE. This motivates us then to consider GEE-assisted model selection methods for GLLVMs, which we accomplish by using information criteria formed from appropriate score and Wald statistics to construct fast forward selection following by pruning. We show under general conditions that this GEE-assisted selection approach is asymptotically consistent for GLLVMs, while simulations studies demonstrate their computational efficiency and competitive finite sample performance.

日時

2019年9月11日(水 Wednesday) 19:00-21:00  ※日時に注意

場所

東京大学大学院経済学研究科 学術交流棟 (小島ホール)2階 小島コンファレンスルーム [地図] ※場所に注意
in Kojima Conference Room on the 2nd floor of the Economics Research Annex (Kojima Hall) [Map]

報告

Alex Lamb (University of Montreal)
"Generators, Manifolds, and Adversarial Mixup Resynthesis"

要旨(Abstract) A motivating idea behind deep learning is that a model can learn to map from the high-dimensional space of observations unto a low-dimensional space of salient explanatory factors which vary across the data. Generative models with latent variables perhaps embody this idea most closely. In this talk we'll explore the relationship between deep models and manifolds, trying to make the relationship more explicit and rigorous. At the same time we'll discuss a recent paper, "Adversarial Mixup Resynthesis", that takes a new perspective on latent variables in generative models.