A New Investment Clock Based On Xgboost Method

This thesis builds a new investment clock,Committee-xgb clock,which based on the Xgboost algorithm.The monthly forecast accuracy and the strategy effect of this clock are both better than Merrill clock.Committee-xgb clock uses some methods to improving the accuracy of classification such as double-tag classification,committee method and contradiction adjustment.These methods make the algorithm of our clock exceed some other ordinary machine learning classification algorithms,such as classification tree,random forest and BP neural network.Committee-xgb clock also incorporates multiple macro factors in training.The macroeconomic information covered by these factors is wider than Merrill’s factors,GDP and CPI.And this thesis draws on the idea of AR model and sets "leading factors".This makes the Committee-xgb clock superior to the simple Merrill Lynch clock in using China’s macro-information.We perform factor analysis on the clock factors to under-stand the specific identification capabilities of each factor for the rotation of major assets in detail.The clock factor analysis believes that the interest rate is an important factor in the rotation of large categories of assets in the Chinese market.This is also consistent with some domestic institutions,such as Huatai Securities,whose researchers proposed an interest-rate-based investment clock perspective.In addition,this thesis also finds some other macro factors that have a discriminating effect on the rotation of assets.The ability to identify factors such as balance of trade and inflation can be complemented by interest rate factors.Committee-xgb clock can be used as a macro-quantitative research tool for FOF investment and large-scale asset allocation,becoming a substitute for Merrill Lynch clocks.This clock can assist the asset the researchers to carry out investment research on asset rotation and index rotation.