基于支持向量回归的通胀预期及对中国菲利普斯曲线的实证研究

发布时间：2018-11-25 16:31

【摘要】：对代理人学习行为的刻画以及通过学习行为分析通胀预期形成过程成为近期宏观金融学特别是货币金融学的前沿研究领域,同时计算机技术与人工智能的发展也为学术研究提供了更新更高效的研究工具——机器学习技术。机器学习技术专门研究计算机怎样模拟或实现人类的学习行为,这使得机器学习技术自然而然地成为研究宏观经济学中的学习型预期或者说公众学习行为的首选工具。在机器学习技术中,支持向量机(SVM)及支持向量回归(SVR)以其较为完备的理论基础以及在解决小样本、非线性及高维模式识别中表现出许多特有的优势,成为目前最为广泛使用的机器学习算法之一。本文所提出的SVR通胀预期借鉴了适应性学习预期通过每期不断纳入新信息来刻画的学习机制,同时对信息根据其获取是否存在滞后性进行划分,再采用的支持向量回归(SVR)算法生成通胀预期,并采用GMM方法对五种不同预期形式的菲利普斯曲线进行实证分析。实证分析结果表明:(1)我国菲利普斯曲线中产出缺口和通货膨胀的权衡机制失效,不同通胀预期项的系数均十分显著,因此中央银行在制定货币政策时,应当注重货币政策的独立性,同时应依据通胀预期来进行相应的预期管理。产出缺口项系数均为负值或不显著,这表明我国以菲利普斯曲线为基础的货币政策传导机制失效,即中央银行无法通过改变产出缺口来调控通胀率。同时不同通胀预期项的系数均十分显著,这表明我国菲利普斯曲线具有典型的预期增广的特征或混合预期增广的特征。(2)SVR预期相对于适应性学习预期是一种更"高级"的预期学习方式。SVR通胀预期比理性预期表现出滞后特征,而比适应性预期表现出先行特征。适应性学习预期以适应性预期为基础,因而学习速度较慢。SVR通胀预期的均值、中位数、标准差和偏度都最小,因而SVR通胀预期相对于理性预期、适应性预期以及以适应性预期为基础的适应性学习预期更为合理,也适宜作为央行制定通胀目标区间的合理选择。(3)我国菲利普斯曲线同时具有SVR通胀预期与理性预期的混合学习特征,且SVR通胀预期特征显著强于理性预期特征。混合学习特征表明我国通胀预期不完全向前看,而是有限理性的,因而货币政策调整时不能完全前瞻,而应随学习预期的不断递归进行微调。SVR通胀预期显著表明,我国菲利普斯曲线不仅具有学习特征,而且公众对通胀预期有区别于适应性学习的更"高级"的学习方式,因而信息获取能力较强,信息维度较高。因此,针对SVR通胀预期,中央银行应从两个方面展开预期管理:一方面,中央银行需要引导公众对通胀的学习行为,应提高货币政策的透明度和信息披露水平,建立货币政策信息平台加强与经济个体的信息沟通,从而使公众尽可能多地掌握学习过程中所需要的信息,加快通胀预期的学习速度;另一方面,中央银行除提高信息沟通效率之外,还应加强与财政政策、汇率政策、产业政策等其他宏观经济政策的政策协调,从而使其他经济变量能够充分及时地反映通胀预期形成的信息。本文的创新之处在于:第一,提出SVR通胀预期以刻画代理人面对存在滞后效应的高维信息样本时的通胀预期形成机制,并使用支持向量回归(SVR)算法估计出SVR通胀预期。在估计过程中,本文创新性地将变量分为存在滞后效应的变量和不存在滞后效应的变量。进一步本文将SVR通胀预期与理性通胀预期和适应性通胀预期进行比较,通过分析三种通胀预期的时序图和统计特征,认为SVR通胀预期更能合理刻画公众对通胀预期的学习行为,从而更适宜作为货币政策通胀目标的选择,这一点为本文实证分析我国不同预期的菲利普斯曲线时,将菲利普斯曲线形式区分为适应性预期的菲利普斯曲线、理性预期的菲利普斯曲线、SVR预期的菲利普斯曲线、适应性预期与理性预期混合的菲利普斯曲线、SVR预期与理性预期混合的菲利普斯曲线提供了基础。第二,本文基于GMM方法,对我国五种不同预期形式的菲利普斯曲线进行实证分析,并依据实证结果对不同预期形式的菲利普斯曲线进行比较,认为SVR通胀预期与理性通胀预期混合的菲利普斯曲线更能刻画我国货币政策的传导机制,同时由于SVR通胀预期更能刻画公众的学习行为,因而通过实证分析混合菲利普斯曲线中SVR预期项系数与理性预期项系数,本文进一步分析了菲利普斯曲线的混合预期中,公众对通胀预期的学习行为相对于理性预期行为的重要性。
[Abstract]:The characterization of the agent's learning behavior and the analysis of the expected formation of inflation through learning behavior have become the frontier research field of recent macro-finance, especially the monetary finance, At the same time, the development of computer technology and artificial intelligence also provides a more efficient research tool _ machine learning technology for academic research. The machine learning technology is a special study on how the computer can simulate or realize the human learning behavior, which makes the machine learning technology become the preferred tool for studying the learning-type expectation or the public learning behavior in the macroeconomics. In the machine learning technology, the support vector machine (SVM) and the support vector regression (SVR) have many unique advantages in solving small samples, non-linearity and high-dimensional pattern recognition, and become one of the most widely used machine learning algorithms. The proposed SVR inflation is expected to draw on the learning mechanism that the adaptive learning is expected to be characterized by the continuous inclusion of new information, and the information is divided according to the existence of the hysteresis, and the support vector regression (SVR) algorithm is used to generate the inflation expectation. The method of GMM is used to analyze the Phillips curve of five different expected forms. The results of the empirical analysis show that (1) The trade-off mechanism of the output gap and inflation in the Phillips curve of our country has failed, and the coefficients of different inflation expectations are all significant, so the central bank should pay attention to the independence of the monetary policy in the formulation of monetary policy. At the same time, the corresponding expected management should be carried out according to the inflation expectations. The coefficient of output gap term is negative or insignificant, which indicates that the monetary policy transmission mechanism based on the Phillips curve has failed, that is, the central bank cannot control the inflation by changing the output gap. At the same time, the coefficients of the different inflation expectations are significant, which suggests that the Phillips curve in our country has a typical expected augmented feature or a mix of expected augmented features. (2) SVR is expected to be a more "high-level"-expected learning approach with respect to adaptive learning. SVR inflation is expected to show a hysteresis characteristic than the rational expectation, and it shows the leading feature more than that of the adaptive expectation. Adaptive learning is expected to be based on an adaptive expectation, so the learning speed is slow. The expected mean, median, standard deviation, and bias of SVR inflation are the smallest, so that SVR inflation is expected to be more reasonable with respect to rational expectations, adaptive expectations, and adaptive learning based on adaptive expectations, as well as a reasonable choice of the central bank to develop an inflation target interval. (3) The Phillips curve of our country also has the mixed learning characteristics of the expected and rational expectation of the SVR inflation, and the expected characteristics of the SVR inflation are significantly stronger than the rational expected characteristics. The mixed learning characteristics show that the inflation expectations of our country are not completely forward, but are limited and rational, so the adjustment of the monetary policy cannot be fully forward, and the fine adjustment should be made with the continuous recursion of learning expectation. SVR inflation is expected to show that the Phillips curve of our country has not only the learning characteristics, but also the public's higher "high-level" of learning that the inflation is expected to be different from the adaptive learning, so the information acquisition ability is strong and the information dimension is high. Thus, for SVR inflation expectations, the central bank should expand its intended management in two ways: on the one hand, the central bank needs to guide the public's learning of inflation and should increase the transparency of monetary policy and the level of information disclosure, To set up a monetary policy information platform to strengthen the information communication with the economic individual, so that the public can master the information needed in the learning process as much as possible, and speed up the expected learning speed of the inflation; on the other hand, the central bank should strengthen the financial policy in addition to improving the information communication efficiency, The policy coordination of other macroeconomic policies, such as the exchange rate policy, the industrial policy, and the like, allows other economic variables to reflect the information expected to be formed in a full and timely manner. The innovation of this paper is that the first, it is expected that SVR inflation is expected to depict the expected formation mechanism of the inflation expectations when the agent faces the high-dimensional information samples with a lag effect, and the SVR inflation expectations are estimated using the support vector regression (SVR) algorithm. In the estimation process, the variable is divided into a variable with a hysteresis effect and a variable which does not have a hysteresis effect. In this paper, SVR inflation is expected to be compared with the expected and adaptive inflation expectations of the rational inflation. By analyzing the timing chart and the statistical feature of the three inflation expectations, it is considered that the SVR inflation is expected to more reasonably characterize the public's expected learning behavior of inflation. Therefore, it is more suitable for the choice of the target of monetary policy inflation. The Phillips curve, which is expected by the SVR, is based on the Phillips curve, which is expected to be mixed with the rational expectation, and SVR is expected to be based on the Phillips curve that is expected to be mixed with the rational expectation. Secondly, based on the GMM method, this paper makes an empirical analysis of the Phillips curves of five different expected forms in China, and compares the Phillips curves in different expected forms according to the empirical results. The Phillips curve, which is thought to be mixed with the expected combination of the inflation of the SVR and the rational inflation, can describe the conduction mechanism of the monetary policy in our country, and at the same time, as the SVR inflation is expected to be more capable of portraying the public's learning behavior, Therefore, through the empirical analysis of the expected coefficient of SVR and the factor of rational expectation in the mixed Phillips curve, this paper further analyses the importance of the public's expected behavior in the expectation of inflation relative to the expected behavior of the rational expectation in the mixed expectation of the Phillips curve.
【学位授予单位】：东北财经大学
【学位级别】：硕士
【学位授予年份】：2016
【分类号】：F822.5

【相似文献】