2011-10-012024-05-17https://scholars.lib.ntu.edu.tw/handle/123456789/687959摘要：由於各國人口老化的趨勢，關於老人退休金的各種制度設計，愈來愈成為政策辯論的焦點。其中，有一些研究嘗試去估計老人的年金所得對他們原來所收到的私人移轉所產生的「排擠效應」。估計這個排擠效應，對於評估年金計畫的政策效果非常關鍵。因為如果排擠效應很顯著，忽略它將會嚴重高估該政策效果。甚至，如果排擠效應達到一比一的關係 (也就是一塊錢的年金排擠掉一塊錢的私人移轉)，那麼年金計畫將達不到任何增加老人所得的效果。 儘管過去嘗試估計這個「排擠效應」的研究使用了不同國家的資料，也採用不同的估計方法，這些研究的估計結果往往都受到變數內生性的影響，而可能有著不同程度的估計誤差。為此，本計畫將使用「回歸不連續設計」(Regression Discontinuity Design, or RD design) 這個方法，來處理變數內生性的問題，以獲得「排擠效應」更可靠的估計值。RD design 這個方法的一個優越性在於，當一個資料生產過程 (data generation process, DGP) 使得某個變數在某個值(即臨界點)的位置呈現出不連續的跳動，而導致這個不連續跳動的原因是外生的，這就使得這個臨界點的左邊(比臨界點略低的點)與右邊(比臨界點略高的點)的結果變數(outcome variable)上的差異，可以被解釋成實驗效果(treatment effect)。這一個特性，使得透過用RD design 估計出來的政策效果相對可靠，比較沒有內生性偏誤的疑慮。甚至在某些情形下，RD design 的估計值，比用其他非實驗性方法(例如最小平方法，工具變數法，或者difference-in-difference 法)得到的估計值，還要更可靠。 本計畫將用台灣的老農津貼作為例子，並用「台灣地區家庭收支調查」的統計資料來進行估計。用這個例子與這些資料的一個優點是，最小平方法與工具變數法也可以應用進來，而分別得到估計值。這些估計值可以拿來與RD 估計值進行比較，以分析各種估計方法的表現。<br> Abstract: Given the aging of the population, policies relating to the design and reform of public pension programs are prominent in policy debates. Previous studies have attempted to measure the extent to which a provision of an old age pension program induced ‘crowding out’ of transfers made by other family members to the elderly. A precise measure of the crowding out primarily constitutes the core of policy evaluation, as ignoring it usually leads to over-statement of policy outcomes. Further, in the case of a complete (dollar to dollar) crowding out, the intended policy impact is entirely neutralized by the private response, leaving the recipients no better off than before financially. While previous studies exploring the crowding out differ in data and strategies employed in the analyses, it is common that estimates from these studies are potentially contaminated by endogeneity due to omitted variables bias and self-selection. To address the endogeneity issue, in this proposed project I am planning to extend the application of regression discontinuity (RD) design to refining the assessment of the degree of crowding out, using Farmers’ Pension Program (FPP) in Taiwan as the experiment. The primary advantage of RD design is that, when the data generation process (DGP) produces a discontinuity in the forcing variable at the threshold point, the requirements for an application of a RD design can be less stringent for points around the threshold – it demands that individuals are unable to precisely manipulate, though still have some influence on, the forcing variable. In this case, the variation in treatment near the threshold can be considered as close to be randomized. This advantage may help overcome some challenges that often threat previous attempts to estimate crowding out using other non-experimental approaches, such as ordinary least squares (OLS), two-stage least squares (2SLS), and difference-in-difference (DD) methods. Further, the experiment of FPP allows estimating the crowding-out effect using both OLS and 2SLS strategies, and the results can be compared to the RD estimate. These comparisons are intriguing because they shed light on whether OLS results are subject to omitted variables bias, and whether the IV strategy can properly address the endogeneity. Further, these results can also be weighed against the DD estimate of the crowding out, which is available in Fan (2010). The comparison, therefore, will provide a unique opportunity to evaluate the validity of applying the DD strategy and the validity of the employed control groups in that strategy.&#63796農津貼私人移轉回歸&#63847&#63898續設計public pensionintervivos transfersregression discontinuity design