為更好地推動(dòng)統(tǒng)計(jì)學(xué)、數(shù)據(jù)科學(xué)及相關(guān)學(xué)科的發(fā)展，促進(jìn)國(guó)內(nèi)青年統(tǒng)計(jì)學(xué)者之間的學(xué)術(shù)交流與合作，全國(guó)工業(yè)統(tǒng)計(jì)學(xué)教學(xué)研究會(huì)青年統(tǒng)計(jì)學(xué)家協(xié)會(huì)2026年年會(huì)將于2026年4月11日在西南財(cái)經(jīng)大學(xué)柳林校區(qū)（成都）舉辦。此次會(huì)議由全國(guó)工業(yè)統(tǒng)計(jì)學(xué)教學(xué)研究會(huì)青年統(tǒng)計(jì)學(xué)家協(xié)會(huì)主辦，西南財(cái)經(jīng)大學(xué)統(tǒng)計(jì)與數(shù)據(jù)科學(xué)學(xué)院、西南財(cái)經(jīng)大學(xué)統(tǒng)計(jì)交叉創(chuàng)新研究院、西南財(cái)經(jīng)大學(xué)數(shù)據(jù)科學(xué)與商業(yè)智能聯(lián)合實(shí)驗(yàn)室承辦，《統(tǒng)計(jì)理論及其應(yīng)用（英文）》編輯部、狗熊會(huì)協(xié)辦。論壇邀請(qǐng)國(guó)內(nèi)外知名統(tǒng)計(jì)學(xué)家和杰出青年統(tǒng)計(jì)學(xué)者做大會(huì)報(bào)告，并邀請(qǐng)國(guó)內(nèi)優(yōu)秀青年統(tǒng)計(jì)學(xué)者到會(huì)開(kāi)展深入探討，也為有志于進(jìn)入高校發(fā)展的統(tǒng)計(jì)學(xué)人才以及有意求賢的高校，提供互相展示、溝通和了解的交流平臺(tái)（點(diǎn)擊閱讀原文下載會(huì)議通知）。
邀請(qǐng)報(bào)告｜復(fù)雜時(shí)間序列分析
報(bào)告題目
Structural Break-driven Optimal Subsample Forecast Combination for Factor-augmented Regressions
報(bào)告人簡(jiǎn)介
汪思韋, 湖南大學(xué)
汪思韋，湖南大學(xué)金融與統(tǒng)計(jì)學(xué)院副教授。主要從事非參數(shù)計(jì)量建模、高維數(shù)據(jù)分析等領(lǐng)域研究。目前在 Journal of Econometrics, Oxford Bulletin of Economics and Statistics，Economics Letters以及《系統(tǒng)工程理論與實(shí)踐》等計(jì)量經(jīng)濟(jì)學(xué)專(zhuān)業(yè)期刊上發(fā)表學(xué)術(shù)論文?，F(xiàn)主持國(guó)家自然科學(xué)基金青年科學(xué)基金項(xiàng)目以及湖南省自然科學(xué)基金青年科學(xué)基金項(xiàng)目。
報(bào)告摘要
In the practice of economic and financial time series forecasting in data-rich environment, structural breaks are pervasive, and while integrating pre- and post-break data has long been recognized to potentially enhance prediction accuracy compared to relying solely on post-break information, a consensus on effectively leveraging break information remains elusive. This paper addresses this gap by proposing a novel subsample forecast combination scheme for factor-augmented regressions: subsamples are constructed based on the identified most recent break, with a subsample tuning parameter governing subsample specifications (length and quantity), candidate forecasts are generated from factor-augmented regressions within each subsample to summarize break-related information (e.g., magnitude, location), and forecast combinations are derived via weights that minimize a forward validation criterion alongside optimal subsample specification selection. Theoretical analysis establishes uniform consistency of estimated coefficients and asymptotic optimality of selected weights and subsample specifications; further, if correctly-specified models exist among candidate subsample forecasts, they are assigned all weights with probability approaching one. Numerical results from simulations and a real-data application to macroeconomic forecasting demonstrate the combination strategy's superior practical performance.
報(bào)告題目
Generative Doubly Robust Estimation for General Treatment Effects
報(bào)告人簡(jiǎn)介
鐘齊先, 廈門(mén)大學(xué)
鐘齊先，廈門(mén)大學(xué)經(jīng)濟(jì)學(xué)院副教授，主要研究領(lǐng)域?yàn)樯鏀?shù)據(jù)、深度學(xué)習(xí)、函數(shù)型數(shù)據(jù)分析和因果推斷等，其學(xué)術(shù)成果發(fā)表在A(yíng)OS、Biometrika、JASA、JBES、JOE、NeurIPS等學(xué)術(shù)期刊或會(huì)議上，主持國(guó)家自然科學(xué)基金青年和面上項(xiàng)目各一項(xiàng)和主要參與一項(xiàng)國(guó)家重點(diǎn)研發(fā)計(jì)劃青年科學(xué)家項(xiàng)目。
報(bào)告摘要
This paper introduces a unified framework for doubly robust (DR) estimation of a broad class of causal functionals, including average, quantile, and asymmetric least squared treatment effects, as well as their conditional counterparts. While DR estimators are well-established for average treatment effects, their development for distributional parameters like quantile treatment effects has not yet been investigated. We bridge this gap by integrating conditional generative models into a loss-based estimating framework. Our approach uses generative models to synthesize counterfactual samples, defines a target loss whose minimizer corresponds to the causal functional of interest, and constructs a final DR estimator by combining these elements with inverse probability weighting. The resulting estimators are shown to be root-$n$ consistent, asymptotically normal, and semiparametrically efficient for unconditional effects, provided either the propensity score or the generative model is correctly specified. For conditional effects, we employ deep neural networks, establishing minimax-optimal convergence rates that adapt to low intrinsic data structures. Simulations confirm the double robustness and finite-sample performance of the proposed methods. This work provides a robust and flexible tool for distributional and heterogeneous causal inference in observational studies, where model misspecification is a persistent concern.
報(bào)告題目
Estimation of Change Points in High-Dimensional Constrained Factor Models with Small T
報(bào)告人簡(jiǎn)介
向鏡潔，華中師范大學(xué)
向鏡潔，華中師范大學(xué)經(jīng)濟(jì)與工商管理學(xué)院副教授，研究方向包括高維因子模型、交互效應(yīng)面板模型的理論與應(yīng)用，以第一作者或通訊作者身份在《Econometric Reviews》、《Oxford Bulletin of Economics and Statistics》、《Economics Letters》、《Pacific-Basin Finance Journal》、《Emerging Markets Review》、《金融研究》等期刊發(fā)表論文多篇，主持國(guó)家自然科學(xué)基金青年項(xiàng)目1項(xiàng)。
報(bào)告摘要
This paper considers the estimation of break points in high-dimensional constrained factor models where the number of time periods (T) is much smaller than the number of cross sections (N). We establish the conditions under which the least squares (LS) estimator is consistent for both large and small breaks. For large breaks, the estimator is also consistent under fixed T. If the number of factors is unobservable, we further show the consistency of the LS estimator based on the estimated number of pseudo factors minus one. Simulation results confirm that the break date can be accurately estimated when T is small. In empirical applications, the method is implemented to estimate the break date in global firm-level carbon emissions from 2010–2019. We find that the structural break occurs in 2015, which coincides with the establishment of the Paris Agreement.
報(bào)告題目
Deep Nonlinear Factor Dimension Reduction for Multiple Time Series via GMDD
報(bào)告人簡(jiǎn)介
戴爽，中國(guó)科學(xué)院
戴爽，中國(guó)科學(xué)院數(shù)學(xué)與系統(tǒng)科學(xué)研究院博士后，2024年博士畢業(yè)于華東師范大學(xué)統(tǒng)計(jì)學(xué)院。主要研究方向?yàn)楦呔S數(shù)據(jù)分析與時(shí)間序列分析，相關(guān)研究成果發(fā)表于Statistica Sinica、Statistics and Computing以及Journal of Nonparametric Statistics等學(xué)術(shù)期刊。
報(bào)告摘要
We propose a novel framework for factor dimension reduction in vector time series by extending the deep nonlinear sufficient dimension reduction method (GMDD-Net; Chen et al., 2024) to strictly stationary processes. The proposed approach aims to construct a contemporaneous nonlinear transformation of a p-dimensional time series into a small number of lower-dimensional subseries, thereby effectively capturing the underlying nonlinear factor structures. From a theoretical perspective, we show that the proposed one-step estimation procedure with squared Frobenius norm regularization is unbiased at the general σ-field level. To establish non-asymptotic convergence rates, we recursively apply the Coupling Lemma of Rio (2017) to construct a sequence of independent blocks of length q0, which allows us to rigorously control the approximation error induced by replacing the initial β-mixing sequence with its coupled version. The resulting convergence rates are slower than those obtained in the i.i.d. case by a factor of (log n)1r for r1 under exponential decay, and by n1r for r>2 under polynomial decay, reflecting the impact of temporal dependence in multivariate time series. Numerical experiments on both simulated and real datasets indicate that the proposed method effectively reduces dimensionality while preserving essential nonlinear structures, facilitating the modeling and forecasting of high-dimensional nonlinear dynamical systems.
博士生論壇
此次會(huì)議設(shè)有超過(guò)10場(chǎng)博士生論壇，歡迎在讀博士生投稿（投稿要求會(huì)在稍后專(zhuān)門(mén)發(fā)布推文）。一旦入選將有機(jī)會(huì)在博士生論壇進(jìn)行宣講，并且得到travel award。成功宣講的博士生會(huì)得到協(xié)會(huì)頒發(fā)的宣講證書(shū)。
高校招聘專(zhuān)場(chǎng)
為了更好地促進(jìn)高校與博士生之間的交流，協(xié)會(huì)特設(shè)高校招聘專(zhuān)場(chǎng)，費(fèi)用6000元/個(gè)。2024年和2025年的年會(huì)成功吸引了超過(guò)30家高校和企業(yè)到現(xiàn)場(chǎng)進(jìn)行招募。有意向的高校請(qǐng)將基本情況發(fā)送郵件到 [email protected]，與李老師聯(lián)絡(luò)。請(qǐng)?jiān)卩]件中說(shuō)明學(xué)?；?qū)W院的基本情況，聯(lián)系人方式等。
會(huì)議舉辦地點(diǎn)
西南財(cái)經(jīng)大學(xué)柳林校區(qū) 四川成都溫江柳臺(tái)大道555號(hào)
聯(lián)系人：潘蕊 [email protected]
掃描下方二維碼即可報(bào)名參會(huì)（本次會(huì)議不收取會(huì)務(wù)費(fèi)，食宿等費(fèi)用自理）
