學術產出-Theses
Article View/Open
Publication Export
-
題名 通貨膨脹目標機制的政策效果估計 : 機器學習與模型平均方法之實證應用
Estimating Treatment Effects of Inflation Targeting with Machine Learning and Model Combination作者 羅婉蘋
LO, Wan-Ping貢獻者 廖仁哲
Liao, Jen-Che
羅婉蘋
LO, Wan-Ping關鍵詞 因果推論
通貨膨脹目標機制
機器學習
模型平均方法
實證蒙地卡羅模擬
Causal Inference
Inflation Targeting
Machine Learning
Model Averaging
Empirical Monte Carlo Simulations日期 2023 上傳時間 2-Aug-2023 13:42:03 (UTC+8) 摘要 本研究旨在應用機器學習與模型平均方法估計通膨目標機制(inflation targeting, IT)的政策效果。我們採用7種機器學習方法估計傾向分數(propensity scores),亦即採行IT政策的條件機率;不同於機器學習模型的傳統作法,我們納入組間變數均衡(covariate balance)、共同支撐區間(common support)等攸關政策效果認定(identification)條件的相關指標,用以調校機器學習相關超參數(hyperparameters),以反映估計「因果」關係,而非「預測」的研究焦點;最後,我們以機率倒數加權法(inverse probability weighting)估計實施IT對於降低通膨率的平均處理效果(average treatment effect, ATE)與處理組平均處理效果(average treatment effect on the treated, ATT)。本文的另一研究重點在於利用模型平均(model averaging)方法合併(或線性組合)前述各種機器學習模型所得之ATE與ATT 估計值,以期進一步改善單一個別機器學習模型的政策效果估計。我們的模型平均方法包括模型簡單平均(Equal Weight, EQ)與模型加權平均(Combined),這兩種模型平均方法皆非常直觀與簡易計算,前者的模型組合權重固定、不需估計,而後者的模型組合權重由資料決定,可用最小平方法估計而得。我們利用實證蒙地卡羅模擬法 (empirical Monte Carlo simulation) 比較前述之機器學習與模型平均方法對於安慰劑ATE/ATT的估計表現,實證結果發現相對於單一個別機器學習模型,Combined與EQ之模型平均方法皆可顯著降低安慰劑ATE/ATT的估計偏誤與變異數,進而改善以均方差(mean squared error) 衡量的整體估計誤差。此外,Combined的估計表現較EQ更為穩定,且隨著模擬樣本數的增加,其相對改善幅度益見顯著。此外,我們於模擬後得出的另一個實證發現在於:相較於以準確率 (accuracy) 為機器學習調參的績效指標,以組間變量平衡、共同支撐及重疊區相關指標進行調參可產生較為穩定之IT政策效果估計值。有關實施IT政策對於降低通膨率的效果估計,我們的主要結果係利用1980 - 2007年60個開發中國家,應用兩種模型平均法所得之降低通膨率的政策效果估計值分別為1.425~1.764百分點(ATE)與3.177~3.909百分點(ATT),後者具1%統計顯著性。我們進一步比較早期與後期(以1999年為界)實施IT的政策效果,結果顯示在前期實施IT政策更能有效降低通膨率,其ATE與ATT估計值分別為0.609~1.006百分點與1.994~2.291百分點,而後期的對應估計值為0.576~0.880百分點與0.882~2.030百分點,此一發現與Bhalla et al. (2023)的實證結果一致。最後,我們變動不同IT認定標準、是否剔除惡性通膨國家、不同資料起始時間、增加變數維度等,前述的實證結果皆具穩健性。
This study applies machine learning (ML) and model averaging methods to estimating the treatment effects of adopting inflation targeting (IT) on lowering inflation levels. We use various machine learning methods to estimate the propensity score---the conditional probability of adopting an IT policy. To reflect our focus on estimating causal effects rather than predicting, as targets of tuning ML hyperparameters, we include covariate balance, common support, and other indicators that are relevant to identification conditions of treatment effects. We then use the inverse probability weighting approach to estimate average treatment effects (ATE) and average treatment effects on the treated (ATT).Another main focus of this study is to use the idea of model averaging to combine the ATE/ATT estimates obtained from our ML methods. Our model averaging methods are simple and easy to implement. Specifically, we consider simple averaging with equal weights (EQ) across models and weighted averaging (Combined) with data-driven weights, which can simply be estimated by least squares.To compare the performance of the ML and model averaging methods, we conduct an empirical Monte Carlo simulation study to estimate the placebo ATE/ATT. Our simulation results reveal that our Combined and EQ methods tend to improve upon the ML approaches in terms of mean-squared errors of the placebo ATE/ATT estimates. Moreover, Combined outperforms EQ in most cases, with more noticeable improvements as the simulation sample size increases. We also find that relative to using accuracy as a ML tuning target, using the aforementioned ATE/ATT identification-related targets can yield more stable ATE/ATT estimates.Lastly, we empirically apply our proposed methods to estimate the ATE/ATT of adopting the IT policy based on a data set consisting of 60 developing countries from 1980 to 2007. Our main results from the proposed Combined and EQ methods show that IT-adoption reduces inflation rates by 1.425-1.764 percentage points (ATE) and 3.177-3.909 percentage points (ATT), with the latter statistically significant at the 1% level. Our results are robust across changing identification standards of IT adoption, whether to exclude countries with hyperinflation, using different data sets, and increasing the dimension of confounding factors.We also find that early IT adopters (pre-1999) enjoyed more success in reducing inflation rates than late adopters, which is consistent with the finding of Bhalla et al. (2023). Specifically, our respective ATE and ATT estimates for early IT-adoption are 0.609-1.006 and 1.994-2.291 percentage points, while the corresponding estimates for late IT-adoption are 0.576-0.880 and 0.882-2.030 percentage points.參考文獻 ALPANDA, S., AND A. HONIG (2014): “The impact of central bank independence on the performance of inflation targeting regimes,” Journal of International Money and Finance,44(C), 118–135.BAJARI, P., D. NEKIPELOV, S. P. RYAN, AND M. YANG (2015): “Demand estimation withmachine learning and model combination,” American Economic Review, 105(5), 481–485.BATINI, N., AND D. LAXTON (2007): “Under what conditions can inflation targeting beadopted?: The experience of emerging markets,” Documentos de Trabajo, (406), 467–506.BERNANKE, B. S., AND F. S. MISHKIN (1997): “Inflation targeting: a new framework formonetary policy?,” Journal of Economic perspectives, 11(2), 97–116.BHALLA, S., K. BHASIN, AND P. LOUNGANI (2023): “Macro Effects of Formal Adoption ofInflation Targeting,” Discussion paper, International Monetary Fund.BLONIARZ, A., H. LIU, C.-H. ZHANG, J. S. SEKHON, AND B. YU (2016): “Lasso adjustmentsof treatment effect estimates in randomized experiments,” Proceedings of the NationalAcademy of Sciences, 113(27), 7383–7390.BREIMAN, L. (2001): “Statistical Modeling: The Two Cultures (with comments and a rejoinder by the author),” Statistical Science, 16(3), 199–231.BRITO, R. D., AND B. BYSTEDT (2010): “Inflation targeting in emerging economies: Panelevidence,” Journal of Development Economics, 91(2), 198–210.BROWN, K., P. MERRIGAN, AND J. ROYER (2018): “Estimating average treatment effectswith propensity scores estimated with four machine learning procedures: simulation results in high dimensional settings and with time to event outcomes,” Available at SSRN:https://ssrn.com/abstract=3272396.CANNAS, M., AND B. ARPINO (2019): “A comparison of machine learning algorithms andcovariate balance measures for propensity score matching and weighting,” BiometricalJournal, 61(4), 1049–1072.CHERNOZHUKOV, V., D. CHETVERIKOV, M. DEMIRER, E. DUFLO, C. HANSEN, W. NEWEY,AND J. ROBINS (2018): “Double/debiased machine learning for treatment and structuralparameters,” Econometrics Journal, 21, 1–68.DUONG, T. H. (2022): “Inflation targeting and economic performance over the crisis: evidence from emerging market economies,” Asian Journal of Economics and Banking,6(3), 337–352.GOLLER, D., M. LECHNER, A. MOCZALL, AND J. WOLFF (2020): “Does the estimation ofthe propensity score by machine learning improve matching estimation? The case ofGermany’s programmes for long term unemployed,” Labour Economics, 65, 101855.GONÇALVES, C. E. S., AND A. CARVALHO (2009): “Inflation targeting matters: Evidencefrom OECD economies’ sacrifice ratios,” Journal of Money, Credit and Banking, 41(1),233–243.GONÇALVES, C. E. S., AND J. M. SALLES (2008): “Inflation targeting in emerging economies:What do the data say?,” Journal of development economics, 85(1-2), 312–318.GOULET COULOMBE, P., M. LEROUX, D. STEVANOVIC, AND S. SURPRENANT (2022): “Howis machine learning useful for macroeconomic forecasting?,” Journal of Applied Econometrics, 37(5), 920–964.GRANGER, C. W., AND R. RAMANATHAN (1984): “Improved methods of combining forecasts,” Journal of forecasting, 3(2), 197–204.GUO, B., H. D. HOLSCHER, L. S. AUVIL, AND WELGE (2021): “Estimating heterogeneoustreatment effect on multivariate responses using random forests,” Statistics in BiosciencesStatistics in Biosciences, 1, 1–17.HASTIE, T., R. TIBSHIRANI, J. H. FRIEDMAN, AND J. H. FRIEDMAN (2009): The elements ofstatistical learning: data mining, inference, and prediction, vol. 2. Springer.HU, Y. (2006): “The choice of inflation targeting—an empirical investigation,” International Economics and Economic Policy, 3, 27–42.HUBER, M., M. LECHNER, AND C. WUNSCH (2013): “The performance of estimators basedon the propensity score,” Journal of Econometrics, 175(1), 1–21.IMAI, K., AND M. RATKOVIC (2013): “Estimating treatment effect heterogeneity in randomized program evaluation,” The Annals of Applied Statistics, 7, 443–470.KING, M. (2002): “The inflation target ten years on,” Available at SSRN:https://ssrn.com/abstract=709004.KREIF, N., AND K. DIAZORDAZ (2019): “Machine learning in policy evaluation: new toolsfor causal inference,” arXiv preprint arXiv:1903.00402.LEE, B. K., J. LESSLER, AND E. A. STUART (2010): “Improving propensity score weightingusing machine learning,” Statistics in medicine, 29(3), 337–346.LIELI, R. P., Y.-C. HSU, AND Á. REGULY (2022): The Use of Machine Learning in TreatmentEffect Estimation. Springer.LIN, S., AND H. YE (2009): “Does inflation targeting make a difference in developing countries?,” Journal of Development Economics, 89(1), 118–123.MCCAFFREY, D. F., G. RIDGEWAY, AND A. R. MORRAL (2004): “Propensity score estimationwith boosted regression for evaluating causal effects in observational studies.,” Psychological methods, 9(4), 403.MCCONNELL, K. J., AND S. LINDNER (2019): “Estimating treatment effects with machinelearning,” Health services research, 54(6), 1273–1282.MENDONÇA, H., AND G. DE GUIMARÃES E SOUZA (2012): “Is inflation targeting a goodremedy to control inflation?,” Journal of Development Economics, 98, 178–191.MOLINA, M., AND F. GARIP (2019): “Machine learning for sociology,” Annual Review ofSociology, 45, 27–45.POSEN, A. S. V., T. LAUBACH, AND F. S. MISHKIN (1998): Inflation targeting: lessons fromthe international experience. Princeton University Press.ROSE, A. K. (2007): “A stable international monetary system emerges: Inflation targeting isBretton Woods, reversed,” Journal of International Money and Finance, 26(5), 663–681.SAMARINA, A., M. TERPSTRA, AND J. DE HAAN (2014): “Inflation targeting and inflationperformance: a comparative analysis,” Applied Economics, 46(1), 41–56.TARR, A., AND K. IMAI (2021): “Estimating average treatment effects with support vectormachines,” arXiv preprint arXiv:2102.11926.TIBSHIRANI, R. (1996): “Regression shrinkage and selection via the lasso,” Journal of theRoyal Statistical Society: Series B (Methodological), 58(1), 267–288.VEGA, M., AND D. WINKELRIED (2005): “Inflation targeting and inflation behavior: a successful story?,” International Journal of Central Banking, 1(3), 153–175.WAHBA,G.,ETAL.(1999): "Supportvectormachines,reproducingkernelHilbertspacesand therandomizedGACV,” Advances inKernelMethods-SupportVectorLearning, 6, 69–87.WILLARD, L.B.(2012):“Doesinflationtargetingmatter?Areassessment,” Applied Eco-nomics, 44(17),2231–2244.YING, X.(2019):“Anoverviewofoverfittinganditssolutions,” Journal ofPhysics:Con-ferenceSeries, 1168,22–23.李怡庭 (2023): “現代央行的信任基礎: 信任–談主要國家抗通膨政策,” 臺灣大學頤賢講座.許育進, 賴宗志 (2018): “處理效果文獻回顧,” 經濟論文叢刊, 46(4),501–521.黃懷德 (2020): “通貨膨脹目標機制政策效果與央行獨立性之實證研究,” 輔仁大學經濟學研究所碩士論文.黃晶晶 (2005): “紐西蘭採行通膨目標化之經驗,” 中央銀行季刊, 27(4),13–18. 描述 碩士
國立政治大學
經濟學系
110258010資料來源 http://thesis.lib.nccu.edu.tw/record/#G0110258010 資料類型 thesis dc.contributor.advisor 廖仁哲 zh_TW dc.contributor.advisor Liao, Jen-Che en_US dc.contributor.author (Authors) 羅婉蘋 zh_TW dc.contributor.author (Authors) LO, Wan-Ping en_US dc.creator (作者) 羅婉蘋 zh_TW dc.creator (作者) LO, Wan-Ping en_US dc.date (日期) 2023 en_US dc.date.accessioned 2-Aug-2023 13:42:03 (UTC+8) - dc.date.available 2-Aug-2023 13:42:03 (UTC+8) - dc.date.issued (上傳時間) 2-Aug-2023 13:42:03 (UTC+8) - dc.identifier (Other Identifiers) G0110258010 en_US dc.identifier.uri (URI) http://nccur.lib.nccu.edu.tw/handle/140.119/146474 - dc.description (描述) 碩士 zh_TW dc.description (描述) 國立政治大學 zh_TW dc.description (描述) 經濟學系 zh_TW dc.description (描述) 110258010 zh_TW dc.description.abstract (摘要) 本研究旨在應用機器學習與模型平均方法估計通膨目標機制(inflation targeting, IT)的政策效果。我們採用7種機器學習方法估計傾向分數(propensity scores),亦即採行IT政策的條件機率;不同於機器學習模型的傳統作法,我們納入組間變數均衡(covariate balance)、共同支撐區間(common support)等攸關政策效果認定(identification)條件的相關指標,用以調校機器學習相關超參數(hyperparameters),以反映估計「因果」關係,而非「預測」的研究焦點;最後,我們以機率倒數加權法(inverse probability weighting)估計實施IT對於降低通膨率的平均處理效果(average treatment effect, ATE)與處理組平均處理效果(average treatment effect on the treated, ATT)。本文的另一研究重點在於利用模型平均(model averaging)方法合併(或線性組合)前述各種機器學習模型所得之ATE與ATT 估計值,以期進一步改善單一個別機器學習模型的政策效果估計。我們的模型平均方法包括模型簡單平均(Equal Weight, EQ)與模型加權平均(Combined),這兩種模型平均方法皆非常直觀與簡易計算,前者的模型組合權重固定、不需估計,而後者的模型組合權重由資料決定,可用最小平方法估計而得。我們利用實證蒙地卡羅模擬法 (empirical Monte Carlo simulation) 比較前述之機器學習與模型平均方法對於安慰劑ATE/ATT的估計表現,實證結果發現相對於單一個別機器學習模型,Combined與EQ之模型平均方法皆可顯著降低安慰劑ATE/ATT的估計偏誤與變異數,進而改善以均方差(mean squared error) 衡量的整體估計誤差。此外,Combined的估計表現較EQ更為穩定,且隨著模擬樣本數的增加,其相對改善幅度益見顯著。此外,我們於模擬後得出的另一個實證發現在於:相較於以準確率 (accuracy) 為機器學習調參的績效指標,以組間變量平衡、共同支撐及重疊區相關指標進行調參可產生較為穩定之IT政策效果估計值。有關實施IT政策對於降低通膨率的效果估計,我們的主要結果係利用1980 - 2007年60個開發中國家,應用兩種模型平均法所得之降低通膨率的政策效果估計值分別為1.425~1.764百分點(ATE)與3.177~3.909百分點(ATT),後者具1%統計顯著性。我們進一步比較早期與後期(以1999年為界)實施IT的政策效果,結果顯示在前期實施IT政策更能有效降低通膨率,其ATE與ATT估計值分別為0.609~1.006百分點與1.994~2.291百分點,而後期的對應估計值為0.576~0.880百分點與0.882~2.030百分點,此一發現與Bhalla et al. (2023)的實證結果一致。最後,我們變動不同IT認定標準、是否剔除惡性通膨國家、不同資料起始時間、增加變數維度等,前述的實證結果皆具穩健性。 zh_TW dc.description.abstract (摘要) This study applies machine learning (ML) and model averaging methods to estimating the treatment effects of adopting inflation targeting (IT) on lowering inflation levels. We use various machine learning methods to estimate the propensity score---the conditional probability of adopting an IT policy. To reflect our focus on estimating causal effects rather than predicting, as targets of tuning ML hyperparameters, we include covariate balance, common support, and other indicators that are relevant to identification conditions of treatment effects. We then use the inverse probability weighting approach to estimate average treatment effects (ATE) and average treatment effects on the treated (ATT).Another main focus of this study is to use the idea of model averaging to combine the ATE/ATT estimates obtained from our ML methods. Our model averaging methods are simple and easy to implement. Specifically, we consider simple averaging with equal weights (EQ) across models and weighted averaging (Combined) with data-driven weights, which can simply be estimated by least squares.To compare the performance of the ML and model averaging methods, we conduct an empirical Monte Carlo simulation study to estimate the placebo ATE/ATT. Our simulation results reveal that our Combined and EQ methods tend to improve upon the ML approaches in terms of mean-squared errors of the placebo ATE/ATT estimates. Moreover, Combined outperforms EQ in most cases, with more noticeable improvements as the simulation sample size increases. We also find that relative to using accuracy as a ML tuning target, using the aforementioned ATE/ATT identification-related targets can yield more stable ATE/ATT estimates.Lastly, we empirically apply our proposed methods to estimate the ATE/ATT of adopting the IT policy based on a data set consisting of 60 developing countries from 1980 to 2007. Our main results from the proposed Combined and EQ methods show that IT-adoption reduces inflation rates by 1.425-1.764 percentage points (ATE) and 3.177-3.909 percentage points (ATT), with the latter statistically significant at the 1% level. Our results are robust across changing identification standards of IT adoption, whether to exclude countries with hyperinflation, using different data sets, and increasing the dimension of confounding factors.We also find that early IT adopters (pre-1999) enjoyed more success in reducing inflation rates than late adopters, which is consistent with the finding of Bhalla et al. (2023). Specifically, our respective ATE and ATT estimates for early IT-adoption are 0.609-1.006 and 1.994-2.291 percentage points, while the corresponding estimates for late IT-adoption are 0.576-0.880 and 0.882-2.030 percentage points. en_US dc.description.tableofcontents 1 緒論 12 文獻回顧 52.1 通膨目標機制介紹 52.2 通膨目標機制的政策效果 62.3 機器學習與因果推論 92.4 集成學習 (ensemble learning) 與模型平均方法 112.5 機器學習與超參數調校 (Hyperparameters Tuning) 123 實證方法 143.1 以機率倒數加權法 (IPW) 為基礎的處理效果 143.2 實證蒙地卡羅模擬 (empirical Monte Carlo simulation) 163.3 機器學習方法 193.3.1 Probit 模型 193.3.2 迴歸及分類隨機森林 193.3.3 羅吉斯最小絕對壓縮挑選機 (logistic least absolute shrinkage and selection operator, Logistic LASSO) 203.3.4 羅吉斯脊迴歸 (Logistic ridge regression) 203.3.5 支撐向量機 (support vector machine, SVM) 213.3.6 類神經網絡 (neural networks) 213.4 模型平均方法 223.4.1 模型加權平均法 (Combined) 223.4.2 模型簡單平均法 (Equal Weight) 233.5 機器學習超參數調校方法244 實證資料 275 實證結果與討論 325.1 實證蒙地卡羅模擬結果 325.1.1 安慰劑處理效果估計 325.1.2 合併處理效果或合併傾向分數? 355.1.3 估計偏誤 (平方) 與變異數分析 365.1.4 改變模擬樣本數 395.1.5 增加干擾變數維度 435.2 ATE 與 ATT 主要估計結果 475.3 前後期實施 IT 政策的處理效果比較 506 穩健性測試 536.1 實施 IT 的時間認定標準 546.2 資料期間涵蓋金融危機事件 556.3 資料剔除惡性通膨的國家 566.4 將 Probit 模型換成羅吉斯 (logistic) 模型 576.5 增加干擾變量維度 586.6 資料涵蓋期間縮減為 1985 - 2007 年 597 結論與建議 607.1 結論 607.2 建議 61參考文獻 62附表 67 zh_TW dc.format.extent 877417 bytes - dc.format.mimetype application/pdf - dc.source.uri (資料來源) http://thesis.lib.nccu.edu.tw/record/#G0110258010 en_US dc.subject (關鍵詞) 因果推論 zh_TW dc.subject (關鍵詞) 通貨膨脹目標機制 zh_TW dc.subject (關鍵詞) 機器學習 zh_TW dc.subject (關鍵詞) 模型平均方法 zh_TW dc.subject (關鍵詞) 實證蒙地卡羅模擬 zh_TW dc.subject (關鍵詞) Causal Inference en_US dc.subject (關鍵詞) Inflation Targeting en_US dc.subject (關鍵詞) Machine Learning en_US dc.subject (關鍵詞) Model Averaging en_US dc.subject (關鍵詞) Empirical Monte Carlo Simulations en_US dc.title (題名) 通貨膨脹目標機制的政策效果估計 : 機器學習與模型平均方法之實證應用 zh_TW dc.title (題名) Estimating Treatment Effects of Inflation Targeting with Machine Learning and Model Combination en_US dc.type (資料類型) thesis en_US dc.relation.reference (參考文獻) ALPANDA, S., AND A. HONIG (2014): “The impact of central bank independence on the performance of inflation targeting regimes,” Journal of International Money and Finance,44(C), 118–135.BAJARI, P., D. NEKIPELOV, S. P. RYAN, AND M. YANG (2015): “Demand estimation withmachine learning and model combination,” American Economic Review, 105(5), 481–485.BATINI, N., AND D. LAXTON (2007): “Under what conditions can inflation targeting beadopted?: The experience of emerging markets,” Documentos de Trabajo, (406), 467–506.BERNANKE, B. S., AND F. S. MISHKIN (1997): “Inflation targeting: a new framework formonetary policy?,” Journal of Economic perspectives, 11(2), 97–116.BHALLA, S., K. BHASIN, AND P. LOUNGANI (2023): “Macro Effects of Formal Adoption ofInflation Targeting,” Discussion paper, International Monetary Fund.BLONIARZ, A., H. LIU, C.-H. ZHANG, J. S. SEKHON, AND B. YU (2016): “Lasso adjustmentsof treatment effect estimates in randomized experiments,” Proceedings of the NationalAcademy of Sciences, 113(27), 7383–7390.BREIMAN, L. (2001): “Statistical Modeling: The Two Cultures (with comments and a rejoinder by the author),” Statistical Science, 16(3), 199–231.BRITO, R. D., AND B. BYSTEDT (2010): “Inflation targeting in emerging economies: Panelevidence,” Journal of Development Economics, 91(2), 198–210.BROWN, K., P. MERRIGAN, AND J. ROYER (2018): “Estimating average treatment effectswith propensity scores estimated with four machine learning procedures: simulation results in high dimensional settings and with time to event outcomes,” Available at SSRN:https://ssrn.com/abstract=3272396.CANNAS, M., AND B. ARPINO (2019): “A comparison of machine learning algorithms andcovariate balance measures for propensity score matching and weighting,” BiometricalJournal, 61(4), 1049–1072.CHERNOZHUKOV, V., D. CHETVERIKOV, M. DEMIRER, E. DUFLO, C. HANSEN, W. NEWEY,AND J. ROBINS (2018): “Double/debiased machine learning for treatment and structuralparameters,” Econometrics Journal, 21, 1–68.DUONG, T. H. (2022): “Inflation targeting and economic performance over the crisis: evidence from emerging market economies,” Asian Journal of Economics and Banking,6(3), 337–352.GOLLER, D., M. LECHNER, A. MOCZALL, AND J. WOLFF (2020): “Does the estimation ofthe propensity score by machine learning improve matching estimation? The case ofGermany’s programmes for long term unemployed,” Labour Economics, 65, 101855.GONÇALVES, C. E. S., AND A. CARVALHO (2009): “Inflation targeting matters: Evidencefrom OECD economies’ sacrifice ratios,” Journal of Money, Credit and Banking, 41(1),233–243.GONÇALVES, C. E. S., AND J. M. SALLES (2008): “Inflation targeting in emerging economies:What do the data say?,” Journal of development economics, 85(1-2), 312–318.GOULET COULOMBE, P., M. LEROUX, D. STEVANOVIC, AND S. SURPRENANT (2022): “Howis machine learning useful for macroeconomic forecasting?,” Journal of Applied Econometrics, 37(5), 920–964.GRANGER, C. W., AND R. RAMANATHAN (1984): “Improved methods of combining forecasts,” Journal of forecasting, 3(2), 197–204.GUO, B., H. D. HOLSCHER, L. S. AUVIL, AND WELGE (2021): “Estimating heterogeneoustreatment effect on multivariate responses using random forests,” Statistics in BiosciencesStatistics in Biosciences, 1, 1–17.HASTIE, T., R. TIBSHIRANI, J. H. FRIEDMAN, AND J. H. FRIEDMAN (2009): The elements ofstatistical learning: data mining, inference, and prediction, vol. 2. Springer.HU, Y. (2006): “The choice of inflation targeting—an empirical investigation,” International Economics and Economic Policy, 3, 27–42.HUBER, M., M. LECHNER, AND C. WUNSCH (2013): “The performance of estimators basedon the propensity score,” Journal of Econometrics, 175(1), 1–21.IMAI, K., AND M. RATKOVIC (2013): “Estimating treatment effect heterogeneity in randomized program evaluation,” The Annals of Applied Statistics, 7, 443–470.KING, M. (2002): “The inflation target ten years on,” Available at SSRN:https://ssrn.com/abstract=709004.KREIF, N., AND K. DIAZORDAZ (2019): “Machine learning in policy evaluation: new toolsfor causal inference,” arXiv preprint arXiv:1903.00402.LEE, B. K., J. LESSLER, AND E. A. STUART (2010): “Improving propensity score weightingusing machine learning,” Statistics in medicine, 29(3), 337–346.LIELI, R. P., Y.-C. HSU, AND Á. REGULY (2022): The Use of Machine Learning in TreatmentEffect Estimation. Springer.LIN, S., AND H. YE (2009): “Does inflation targeting make a difference in developing countries?,” Journal of Development Economics, 89(1), 118–123.MCCAFFREY, D. F., G. RIDGEWAY, AND A. R. MORRAL (2004): “Propensity score estimationwith boosted regression for evaluating causal effects in observational studies.,” Psychological methods, 9(4), 403.MCCONNELL, K. J., AND S. LINDNER (2019): “Estimating treatment effects with machinelearning,” Health services research, 54(6), 1273–1282.MENDONÇA, H., AND G. DE GUIMARÃES E SOUZA (2012): “Is inflation targeting a goodremedy to control inflation?,” Journal of Development Economics, 98, 178–191.MOLINA, M., AND F. GARIP (2019): “Machine learning for sociology,” Annual Review ofSociology, 45, 27–45.POSEN, A. S. V., T. LAUBACH, AND F. S. MISHKIN (1998): Inflation targeting: lessons fromthe international experience. Princeton University Press.ROSE, A. K. (2007): “A stable international monetary system emerges: Inflation targeting isBretton Woods, reversed,” Journal of International Money and Finance, 26(5), 663–681.SAMARINA, A., M. TERPSTRA, AND J. DE HAAN (2014): “Inflation targeting and inflationperformance: a comparative analysis,” Applied Economics, 46(1), 41–56.TARR, A., AND K. IMAI (2021): “Estimating average treatment effects with support vectormachines,” arXiv preprint arXiv:2102.11926.TIBSHIRANI, R. (1996): “Regression shrinkage and selection via the lasso,” Journal of theRoyal Statistical Society: Series B (Methodological), 58(1), 267–288.VEGA, M., AND D. WINKELRIED (2005): “Inflation targeting and inflation behavior: a successful story?,” International Journal of Central Banking, 1(3), 153–175.WAHBA,G.,ETAL.(1999): "Supportvectormachines,reproducingkernelHilbertspacesand therandomizedGACV,” Advances inKernelMethods-SupportVectorLearning, 6, 69–87.WILLARD, L.B.(2012):“Doesinflationtargetingmatter?Areassessment,” Applied Eco-nomics, 44(17),2231–2244.YING, X.(2019):“Anoverviewofoverfittinganditssolutions,” Journal ofPhysics:Con-ferenceSeries, 1168,22–23.李怡庭 (2023): “現代央行的信任基礎: 信任–談主要國家抗通膨政策,” 臺灣大學頤賢講座.許育進, 賴宗志 (2018): “處理效果文獻回顧,” 經濟論文叢刊, 46(4),501–521.黃懷德 (2020): “通貨膨脹目標機制政策效果與央行獨立性之實證研究,” 輔仁大學經濟學研究所碩士論文.黃晶晶 (2005): “紐西蘭採行通膨目標化之經驗,” 中央銀行季刊, 27(4),13–18. zh_TW