| dc.contributor | 國貿系 | |
| dc.creator (作者) | 郭炳伸 | |
| dc.date (日期) | 2022-10 | |
| dc.date.accessioned | 7-Apr-2026 13:28:45 (UTC+8) | - |
| dc.date.available | 7-Apr-2026 13:28:45 (UTC+8) | - |
| dc.date.issued (上傳時間) | 7-Apr-2026 13:28:45 (UTC+8) | - |
| dc.identifier.uri (URI) | https://ah.lib.nccu.edu.tw/item?item_id=181952 | - |
| dc.description.abstract (摘要) | 模型平均的研究大多集中於降低預測風險，很少發展推論理論。本三年計劃用來填補這個空缺。部分問題源於，模型中存在著不能估計的參數。另外一些理論結果，只能應用於巢狀模型，不夠一般。本研究計劃希望提供方法得以處理前述的問題。我們針對參數不能估計提供重複抽樣的建議。對於應用侷限，我們主張應用在高維度模型文化獻中討論到的稀疏性條件，重新檢視既有理論。在研究過程當中，我們將不斷對這些創新想法，分別以理論證明與模擬分析，驗證它們的正確性。 | |
| dc.description.abstract (摘要) | Most of econometric studies on model averaging aim to lower prediction risks. Only very few develops inference theory post model averaging. The 3-year project is to fill the gap. Some part of the problem is due to the existence of an unknown local parameter that generally can not be consistently estimated. Analyzing the model averaging estimator in a local parameter setup in the first place, however,was to help quantify the prediction risks. Alternatively, other authors study the inference post averaging problem in a fixed parameter context. The asymptotic results established are only valid for the nested models. Nevertheless, in practice, many candidate models belong to the family of non-nested models. The inference problem is compounded further by two important observations, possible high-dimensional candidate models and sparse model weighting.The project considers the inference problem in either environment, leaving aside the debates over which setup fits better to realism. We propose two different re-sampling schemes, the residual and wild bootstraps, when in the local parameter context. Both are devised to be free of the local parameter. The proposed residual bootstrap performs well when the candidate models' number is mild.The suggested wild bootstrap is introduced for the high dimension cases. Some preliminary simulation evidence shows promises of our proposed bootstraps.In the fixed parameter framework, the inference problem is more challenging to resolve. We notice the equivalence of the weight selections for the model averaging estimator to the coefficient estimates in a LASSO regression. The correspondence gives rise to the Karush-Kuhn-Tuckeror the sparsity condition, asking for the sources of sparsity. Our asymptotic analysis here will be based on the sparsity condition that is overlooked previously. The project will thoroughly analyze and examine the validity of the ideas both theoretically and numerically. | |
| dc.format.extent | 116 bytes | - |
| dc.format.mimetype | text/html | - |
| dc.relation (關聯) | 科技部, MOST109-2410-H004-126, 109.08-110.07 | |
| dc.subject (關鍵詞) | 平均估計式; 平均後統計推論; 群組; 重複抽樣; 權重選擇; 稀疏性; 預測風險; 高維度 | |
| dc.subject (關鍵詞) | prediction risk; model averaging estimator; residual bootstrap; wild bootstrap;inference post averaging; LASSO; weight selection; sparsity; high dimension | |
| dc.title (題名) | 模型平均後之统計推論 | |
| dc.title (題名) | Inference Post Model Averaging via Bootstrap and Lasso | |
| dc.type (資料類型) | report | |
