朱慧明++歐陽(yáng)文靜++游萬(wàn)海??

摘 要：運(yùn)用貝葉斯動(dòng)態(tài)面板數(shù)據(jù)模型考量中國(guó)出口貿(mào)易的地區(qū)結(jié)構(gòu)差異性，結(jié)果表明：基于Gibbs抽樣算法的貝葉斯動(dòng)態(tài)面板回歸模型能有效刻畫(huà)各地區(qū)出口競(jìng)爭(zhēng)力的路徑依賴(lài)特征；FDI對(duì)出口競(jìng)爭(zhēng)力的影響具有一定的時(shí)滯性，而匯率波動(dòng)對(duì)出口競(jìng)爭(zhēng)力的影響則具有明顯的地域差異性。

關(guān)鍵詞： 出口貿(mào)易；動(dòng)態(tài)隨機(jī)效應(yīng)；貝葉斯分析；FDI；匯率波動(dòng)

中圖分類(lèi)號(hào)：F746.12；O212.8 文獻(xiàn)標(biāo)識(shí)碼： A 文章編號(hào)：1003-7217（2014）02-0109-07

一、引 言

面板數(shù)據(jù)由Mundlak引入到計(jì)量經(jīng)濟(jì)學(xué)研究中，結(jié)合了截面數(shù)據(jù)和時(shí)間序列數(shù)據(jù)的優(yōu)點(diǎn)，通過(guò)截距項(xiàng)刻畫(huà)個(gè)體差異在數(shù)據(jù)調(diào)整過(guò)程中的動(dòng)態(tài)變化，有效減少了數(shù)據(jù)生成過(guò)程中由于加總產(chǎn)生的偏誤，充分利用更多數(shù)據(jù)的信息，提高參數(shù)估計(jì)的有效性和準(zhǔn)確性[1-2]。Balestra和Nerlove發(fā)現(xiàn)大量經(jīng)濟(jì)變量表現(xiàn)出動(dòng)態(tài)滯后效應(yīng)，即經(jīng)濟(jì)變量數(shù)據(jù)除了受當(dāng)期因素的影響，還會(huì)受到非本期因素的影響。為刻畫(huà)動(dòng)態(tài)滯后效應(yīng)，在靜態(tài)面板數(shù)據(jù)模型中引入滯后被解釋變量，構(gòu)建動(dòng)態(tài)面板數(shù)據(jù)模型。該模型可以用于描述多個(gè)經(jīng)濟(jì)變量之間的動(dòng)態(tài)關(guān)系，被廣泛應(yīng)用于金融、經(jīng)濟(jì)、管理等領(lǐng)域[3]。Egger P建立動(dòng)態(tài)面板數(shù)據(jù)模型發(fā)現(xiàn)雙邊貿(mào)易和FDI之間的關(guān)系互補(bǔ)[4-5]。Chiara認(rèn)為垂直型FDI把生產(chǎn)的不同階段放到不同國(guó)家，由此帶動(dòng)中間投入品的進(jìn)口和產(chǎn)成品的出口[6]。Marc Auboin用最大似然估計(jì)法發(fā)現(xiàn)，匯率的不確定性對(duì)出口有顯著的負(fù)面影響[7]。Tang Kin Boon用異質(zhì)面板協(xié)整檢驗(yàn)驗(yàn)證了出口需求函數(shù)的變量之間的協(xié)整關(guān)系，研究發(fā)現(xiàn)外債對(duì)出口的影響因貨幣貶值的幅度而改變[8，9]。Ciarret構(gòu)建面板數(shù)據(jù)模型研究發(fā)現(xiàn)能源價(jià)格和GDP電力消費(fèi)有雙向因果關(guān)系[10]。為揭示經(jīng)濟(jì)學(xué)理論的動(dòng)態(tài)關(guān)系，一些學(xué)者研究動(dòng)態(tài)面板數(shù)據(jù)模型的參數(shù)估計(jì)問(wèn)題，提出三種主要的參數(shù)估計(jì)方法：廣義矩（Generalized Methods of Moments，GMM）方法，校正的最小二乘虛擬變量估計(jì)法（Least Squares with Dummy Variable，LSDV）和層次貝葉斯估計(jì)法。Anderson 使用工具變量方法對(duì)含有一階差分變量的動(dòng)態(tài)面板模型進(jìn)行估計(jì)，得到了模型參數(shù)的一致性估計(jì)[11]。但工具變量法在估計(jì)時(shí)忽視了隨機(jī)誤差項(xiàng)的結(jié)構(gòu)，因此估計(jì)量不具有有效性。Holtz-Eakin在Anderson工具變量估計(jì)法的基礎(chǔ)上，研究了時(shí)變參數(shù)的向量自回歸模型的參數(shù)估計(jì)問(wèn)題[12]。Arrelano將一階差分廣義矩（GMM）估計(jì)方法引入到動(dòng)態(tài)面板數(shù)據(jù)的估計(jì)中[13]。Maurice用局內(nèi)變量或者前定變量的滯后值作為工具變量來(lái)估計(jì)參數(shù)[14]，提高了模型參數(shù)估計(jì)的有效性。HsinChen研究GMM估計(jì)量的大樣本性質(zhì)，發(fā)現(xiàn)當(dāng)動(dòng)態(tài)面板數(shù)據(jù)模型中存在高度序列自相關(guān)性時(shí)，GMM估計(jì)量是有偏差的[15]。Hahn采用校正的最小二乘虛擬變量估計(jì)量去估計(jì)參數(shù)，發(fā)現(xiàn)當(dāng)時(shí)間維數(shù)比較小時(shí)，校正的最小二乘虛擬變量估計(jì)量比GMM方法更精確[16]。Giovanni針對(duì)模型存在的個(gè)體異質(zhì)性提出廣義最小二乘法，并通過(guò)蒙特卡洛研究發(fā)現(xiàn)廣義最小二乘估計(jì)有最小的偏差和均方根誤差[17]。

貝葉斯統(tǒng)計(jì)推斷技術(shù)特別是馬爾科夫鏈蒙特卡洛（MCMC）穩(wěn)態(tài)模擬技術(shù)的發(fā)展，為動(dòng)態(tài)面板數(shù)據(jù)模型的研究提供有效的途徑[18，19]。Hsiao 指出，使用層次貝葉斯方法估計(jì)隨機(jī)效應(yīng)自回歸面板數(shù)據(jù)模型時(shí)，在大面板數(shù)據(jù)條件下，貝葉斯估計(jì)量與組平均估計(jì)量漸進(jìn)等價(jià)。貝葉斯分析方法在動(dòng)態(tài)面板數(shù)據(jù)模型參數(shù)估計(jì)的應(yīng)用，可以避免GMM估計(jì)方法和最小二乘虛擬變量估計(jì)法存在的參數(shù)估計(jì)不準(zhǔn)確、有偏的問(wèn)題[20]。

本文將利用貝葉斯統(tǒng)計(jì)推斷理論，構(gòu)建含外生變量的動(dòng)態(tài)隨機(jī)效應(yīng)面板模型，通過(guò)參數(shù)的分層先驗(yàn)分布研究貝葉斯模型參數(shù)的后驗(yàn)分布（posterior distribution），設(shè)計(jì)相應(yīng)的馬爾科夫蒙特卡洛（Markov chain Monte Carlo，簡(jiǎn)稱(chēng)MCMC）抽樣算法進(jìn)行模型參數(shù)估計(jì)，并對(duì)中國(guó)出口貿(mào)易地區(qū)結(jié)構(gòu)的差異性進(jìn)行實(shí)證分析。

二、貝葉斯動(dòng)態(tài)面板數(shù)據(jù)模型構(gòu)建

（一）模型結(jié)構(gòu)

動(dòng)態(tài)面板模型是一類(lèi)重要的經(jīng)濟(jì)計(jì)量模型，其數(shù)學(xué)表達(dá)形式如下：

四、結(jié) 論

本文應(yīng)用動(dòng)態(tài)面板數(shù)據(jù)模型研究我國(guó)地區(qū)出口貿(mào)易結(jié)構(gòu)差異性，結(jié)果表明，出口貿(mào)易競(jìng)爭(zhēng)力具有很強(qiáng)的路徑依賴(lài)特征。因此要縮小地區(qū)發(fā)展差距需要一個(gè)長(zhǎng)期的過(guò)程，需要不斷引導(dǎo)外商直接投資向中西部地區(qū)轉(zhuǎn)移，加強(qiáng)和完善中西部地區(qū)的優(yōu)惠政策和基礎(chǔ)設(shè)施建設(shè)。同時(shí)，F(xiàn)DI的流入在促進(jìn)出口時(shí)存在一定的時(shí)滯性，即FDI的流入須經(jīng)過(guò)早期投入和生產(chǎn)過(guò)程后才能促進(jìn)出口。從地域差別的角度來(lái)看，F(xiàn)DI的流入加劇了地區(qū)出口貿(mào)易發(fā)展的不平衡，說(shuō)明東部、中部和西部地區(qū)出口貿(mào)易競(jìng)爭(zhēng)力對(duì)人民幣匯率波動(dòng)的敏感性存在顯著差異，即越開(kāi)放的地區(qū)對(duì)人民幣匯率波動(dòng)越敏感。

參考文獻(xiàn)：

[1]Rahman， Sajjadur & Serletis， Apostolos. The effects of exchange rate uncertainty on exports [J]. Journal of Macroeconomics， 2009， 31（3）：500-507.

[2]Rfat Bar Tekin. Economic growth， exports and foreign direct investment in least developed countries：a panel granger causality analysis [J]. Economic Modelling， 2012， 29（3）： 868-878.

[3]Balestra P， M Nerlove. Pooling cross section and time series data in the estimation of a dynamie model： the demand for natural Gas [J]. Econometrica， 1966， 34（3）： 585-612.

[4]Egger P. European exports and outward foreign direct investment： a dynamic panel data approach [J]. Review of World Economics， 2009， 37（3）： 427-449.

[5]蔣仁愛(ài)， 馮根福. 貿(mào)易、FDI、無(wú)形技術(shù)外溢與中國(guó)技術(shù)進(jìn)步 [J]. 管理世界， 2012， 15（9）： 49-60.

[6]Chiara Franco. Exports and FDI motivations：empirical evidence from US foreign subsidiaries [J]. International Business Review， 2013， 22（4）： 47-62.

[7]Marc Auboin， Michel Ruta. The relationship between exchange rates and international trade： a literature review [J]. World Trade Review， 2012， 12（3）： 60-77.

[8]Tang Kin Boon， Tan Hui Boon. The effect of foreign currency borrowing and financial development on exports： a dynamic panel analysis on asiapacific countries [J]. Journal of the Asia Pacific Economy， 2012， 18（3）： 460-476.

[9]Tsay R S， Ando T. Bayesian panel data analysis for exploring the impact of subprime financial crisis on the US stock market [J]. Computational Statistics and Data Analysis， 2012， 56（11）： 3345-3365.

[10]A Ciarret， A Zarraga. Economic growthelectricity consumption causality in 12 european countries： a dynamic panel data approach [J]. Energy Policy， 2010， 38（7）： 3790-3796.

[11]Anderson T W， Hsiao C. Formulation and estimation of dynamic models using panel data [J]. Joumal of Econometrics， 1982， 18（3）： 47-82.

[12]Holtz Eakin D W， Neweyand H S. Estimating vector autoregressions with panel data [J]. Econometrica， 1988， 56（10）： 1371-1395.

[13]Arrelano M， S R Bond. Another look at the instrumental variable estimation of errorcomponents models [J]. Journal of Econometrics， 1995， 68（3）： 29-51.

[14]Maurice J G， Frank Windmeijer. The weak instrument problem of the system GMM estimator in dynamic panel data models [J]. The Econometrics Journal， 2010， 13（1）： 95-126.

[15]Hsin Chen Chang， BwoNung Huang， Chin Wei Yang. Military expenditure and economic growth across different groups：a dynamic panel grangercausality approach [J]. Economic Modelling， 2011， 28（6）： 416-423.

[16]Hahn J， Kuersteiner G. Asymptotically unbiased inference for a dynamic panel model with fixed effects when both n and T are large [J]. Econometrica， 2002， 70（5）： 639-657.

[17]Giovanni S， Bruno. Approximating the bias of the LSDV estimator for dynamic unbalanced panel data models [J]. Economics Letters， 2005， 87（3）： 361-366.

[18]李鯤鵬， 文益俊. 交互效應(yīng)面板模型EM算法和MCMC算法 [J]. 數(shù)理統(tǒng)計(jì)與管理， 2012， 29（4）： 150-161.

[19]吳俊， 賓建成. 中國(guó)商業(yè)銀行操作風(fēng)險(xiǎn)損失分布甄別與分析： 基于貝葉斯MCMC頻率方法 [J]. 財(cái)經(jīng)理論與實(shí)踐，2011，32（5）：8-14.

[20]Cheng Hsiao， A K Tahmiscioglu. Estimation of dynamic panel data models with both individual and timespecific effects [J]. Journal of Statistical Planning and Inference， 2008， 138（9）： 698-721.

（責(zé)任編輯：姚德權(quán)）

Regional Structure of Chinese Export Trade Based on Bayesian Dynamic Panel Data Model

ZHU Huiming，OU YANG Wenjing，YOU Wanhai

. （School of Business Administration，Hunan University，Changsha 410082，China）.

Abstract：This paper constructs a Bayesian dynamic panel data model to address uncertain risk of inference in random coefficient， sets the corresponding prior distribution， and uses the Bayesian theorem to infer the posterior distribution of the parameters. In order to investigate the regional structure of China's export trade， we calculate the indicators export competitiveness （ECI） to measure the regional structure of China's export trade， and then conduct an empirical analysis. The results show that the Bayesian dynamic panel regression model based on Gibbs sampling algorithm can effectively capture the characteristics of the path dependence in export competitiveness of all regions. In addition， the effect of the FDI on the export competitiveness will have a lag， while the impact of the exchange rate fluctuations on the export competitiveness has obvious regional differences.

Key words：Export Trade；Dynamic Random Effects；Bayesian Analysis；Foreign Direct Investment （FDI）；Exchange Rate Fluctuation

[20]Cheng Hsiao， A K Tahmiscioglu. Estimation of dynamic panel data models with both individual and timespecific effects [J]. Journal of Statistical Planning and Inference， 2008， 138（9）： 698-721.

（責(zé)任編輯：姚德權(quán)）

Regional Structure of Chinese Export Trade Based on Bayesian Dynamic Panel Data Model

ZHU Huiming，OU YANG Wenjing，YOU Wanhai

. （School of Business Administration，Hunan University，Changsha 410082，China）.

Abstract：This paper constructs a Bayesian dynamic panel data model to address uncertain risk of inference in random coefficient， sets the corresponding prior distribution， and uses the Bayesian theorem to infer the posterior distribution of the parameters. In order to investigate the regional structure of China's export trade， we calculate the indicators export competitiveness （ECI） to measure the regional structure of China's export trade， and then conduct an empirical analysis. The results show that the Bayesian dynamic panel regression model based on Gibbs sampling algorithm can effectively capture the characteristics of the path dependence in export competitiveness of all regions. In addition， the effect of the FDI on the export competitiveness will have a lag， while the impact of the exchange rate fluctuations on the export competitiveness has obvious regional differences.

Key words：Export Trade；Dynamic Random Effects；Bayesian Analysis；Foreign Direct Investment （FDI）；Exchange Rate Fluctuation

[20]Cheng Hsiao， A K Tahmiscioglu. Estimation of dynamic panel data models with both individual and timespecific effects [J]. Journal of Statistical Planning and Inference， 2008， 138（9）： 698-721.

（責(zé)任編輯：姚德權(quán)）

Regional Structure of Chinese Export Trade Based on Bayesian Dynamic Panel Data Model

ZHU Huiming，OU YANG Wenjing，YOU Wanhai

. （School of Business Administration，Hunan University，Changsha 410082，China）.

Abstract：This paper constructs a Bayesian dynamic panel data model to address uncertain risk of inference in random coefficient， sets the corresponding prior distribution， and uses the Bayesian theorem to infer the posterior distribution of the parameters. In order to investigate the regional structure of China's export trade， we calculate the indicators export competitiveness （ECI） to measure the regional structure of China's export trade， and then conduct an empirical analysis. The results show that the Bayesian dynamic panel regression model based on Gibbs sampling algorithm can effectively capture the characteristics of the path dependence in export competitiveness of all regions. In addition， the effect of the FDI on the export competitiveness will have a lag， while the impact of the exchange rate fluctuations on the export competitiveness has obvious regional differences.

Key words：Export Trade；Dynamic Random Effects；Bayesian Analysis；Foreign Direct Investment （FDI）；Exchange Rate Fluctuation