Research On The Prediction Of Transfer Discount Interest Rates Based On Machine Learning

In recent years,China has gradually accelerated the process of the interest rate liberalization.The efficiency of interest rate transmission has increased as well as the volatility and banks face greater interest rate risk With the improvement of the legal system and the establishment of infrastructure,the Commercial Paper Market in China has become an important sub-market of the monetary market.With the advantages of strong liquidity and good stability,commercial paper is favored by more and more investors.Commercial paper transfer discount business has gradually become an important source of banks' profit.However,with the volatility of overall interest rate increasing,the transfer discount interest rates with a higher level of liberalization fluctuates more and more frequently,and the risks of it become more prominent.As commercial banks are the main traders in the Commercial Paper Market,the fluctuation of the transfer discount interest rates affects the cost of capital,the yield of investment and the allocation of credit lines.Therefore,predicting the transfer discount interest rates more accurately is of great significance to commercial banks in estimating the cost of capital,increasing operating profit,avoiding interest rate risks,allocating credit lines,and balancing internal contradictions.For a long time,Chinese scholars' researches on the pricing of the transfer discount interest rates are mainly theoretical analysis.Empirical researches mainly use traditional econometric models,such as VAR,VECM,etc.This paper attempts to use machine learning methods to analyze it theoretically and empirically,establish a forecast model of the transfer discount interest rates,compare the different factors of the transfer discount interest rates in different period,and put forward the optimal forecast model.Firstly,this paper analyzes the practical significance of the forecast of the transfer discount interest rates for commercial banks,introduces the development of the transfer discount business and analyzes the fluctuation of current transfer discount interest rates.Then,this paper explores the factors that affect the transfer discount interest rates.Macro factors include economic growth,inter-bank funds,credit and the improvement of Commercial Paper Market;Micro factors include the credit rating of the commercial paper,the credit rating of the trade subject,and the denomination of the commercial paper.Then,this paper describes the theories and features of the three methods of Random Forest,eXtreme Gradient Boosting and Support Vector Machine which belong to machine learning methods.Then,this paper introduces the current basic methods of the prediction of the transfer discount interest rates and the forecasting evaluation indicators.Then,this paper analyzes the advantages of machine learning methods in predicting interest rates.Lastly,this paper selects the transfer discount interest rates from January 2010 to July 2019 as the target data set and selects the lagged transfer discount interest rates,the discount interest rates,inter-bank lending rate,seven-day pledged bond repurchase rate,the commercial paper proportion in credit and other indexes as features.Then,Random Forest,eXtreme Gradient Boosting and Support Vector Machine are trained.In order to meet the commercial bank's demand on the transfer discount interest rates in different periods,compare the feature importance in different periods,this paper forecasts the weekly and the monthly transfer discount interest rates respectively.This paper draws the following conclusions: Firstly,the Commercial Paper market interest rates,the commercial paper proportion in credit,other currency market yields have a greater impact on the transfer discount interest rates and the yields of other financial markets and the improvement of Commercial Paper Market have little impact.Secondly,commercial banks should pay different attention to various factors in different periods.Commercial banks should pay more attention to the impact of other currency market yields when predicting the weekly transfer discount interest rates.While commercial banks need to reduce concentrations on the yields of other financial markets and increase concentrations on discount interest rates when forecast the monthly ones.Thirdly,the prediction results indicate: the Support Vector Machine precision is the best for prediction weekly,the Random Forest best monthly.The forecast precision is 93.51% and 83.39% respectively.The forecast precision of three models on the weekly prediction performs significantly better than the monthly.Therefore,commercial banks may choose Support Vector Machine to predict the weekly transfer discount interest rates and Random Forest monthly.