A Study On The Relationship Of The Term Structure Of Interest Rates And The Macro-economy With Regime-Switching

As interest rate marketization reform completed preliminary, the macro-control of China's finance will gradually change from quantitative monetary policy to be price-based monetary policy. The latent macro-economic information in yield curve is of important theoretical and practical significance to forward-looking monetary policy and timely feedback of policy implementation effects. However, most of the existing related literatures only reveal linear relationship of the term structure of interest rates and macroeconomic factors in temporal dimension while ignoring the fact that China is a transition and developing country, the macro economy faces double impact of institutional change and structure adjustment, meanwhile, it is under the background of accelerating opening to the outside world. Since 2000, due to the change of political and economic situation at home and abroad, especially the huge impact of the financial crisis, China's macro-economic fluctuation has intensified. The co-movement of the macro variables and the yield curve is not only time-varying and structure changes but also manifests time-frequency characteristics of periodic fluctuation, and its intensity, direction, synchronization and frequency all change a lot. Based on all of these, we seize the feature by introducing Markov regime-switching factor to study the relationship of the term structure of interest rates and macroeconomic, the main research content is as follows:First of all, we analyze the risk characteristics of the yield curve before and after the financial crisis by principal component analysis. It shows that yield curve mainly shifts un-parallel before the financial crisis while moves parallel after that. It has important enlightenment for commercial banks'interest rate risk management:the choice of risk management tool depends on macroeconomic situation, for the risk of interest rate curve moving parallel, the traditional duration-convexity should be enough. For the risk of interest rate curve shifting un-parallel, we put forward using NS duration vector to immunize it. This research makes up for the defect that the existing domestic literatures only focus on tool itself while lack of considering real macroeconomic condition, this provides empirical support for the set of regime-switching condition in our models.Secondly, taking the reform of exchange rate formation mechanism (July 21,2005) and the launch of short-term liquidity adjustment tool (SLO) (January 18,2013) as time nodes to study the policy impacts on the term structure of interest rates with event analysis. The empirical results show that the yield curve has structure breakpoints before and after the reform, and there exists overreaction for the SLO launch. Based on this, we use the wavelet to study the correlation of fluctuations in the main macro variables and the yield curve from two dimensions of time and frequency. Statistical results show that the level of yield curve contains inflation expectation; the slope can't be used to forecast economic growth, however, it can be taken as an indicator reflecting monetary policy state. In addition, we find that the macroeconomic volatility information in the yield curve appears structure mutation before and after the financial crisis. This study makes up for the inadequacy of existing literatures confined to temporal dimension, and further confirms yield curve-macro economy correlation depends on economy state.Thirdly, on the basis of simple linear models (VAR and DRA), we introduce Markov regime-switching factor to establish compact state-shift macro-finance model (MS-VAR and MSV-DRA). Based on MS-VAR model we find that the term structure-macroeconomic relationship differs significantly in different regimes, these two regimes are closely related to economic growth state; By testing on the MSV-DRA model we find there exists significant two-way influence, impulse response is obvious different in different regimes, variance decomposition shows macro factors can explain more about yield predictive variance in high volatility than in low volatility, it reflects macro variables are necessary for yield prediction; By a Logit model and Taylor rule regression we testify that the high fluctuation and low fluctuation regimes can be interpreted as loose state and tight state of monetary policy respectively. In terms of yield fitting and state-depend tests, these models built describe the term structure-macroeconomic dynamic relation more accurate than simple linear models, and further strengthen its state-shift characteristic.Finally, we compare the out-of-sample yield prediction accuracy of the built compact regime-switching macro-finance models with common single regime models, the prediction evaluation criterions includes not only statistical loss function but also economic loss function. By prediction accuracy comparison it is found that:the direct prediction effect for medium-and long-term yields based on independent Minnesota-Wishart Bayesian conjugate prior distribution vector autoregressive model have obvious advantages among single regime models; While in the comparison among regime-switching models and single regime models, MS-VAR model is not as good as VAR model, MSV-DRA model's medium-and long-term forecast effect is better than DRA model, but its short-term prediction effect is poorer, this result reflects that although the regime-switching factor can improve the model's flexibility and fitting, it is still difficult to obtain high accuracy for China's short sample; Furthermore, the prediction accuracy of DNS model is higher than DRA model shows that macro factors can help to improve prediction, while the over-fitting problem weakens the model's prediction ability. Thus, the flexibility and stability should be balanced when constructing prediction models.