Based On The Functional Data Analysis Of The Futures Price Curve Shape Recognition

Continuously trading in market always updates the price with high frequency. Thetraditional time series analysis method often performs poor when solving financial data, thereason is mainly because the stochastic process of financial data is generally unstable.Functional data analysis treats the observation as a whole, and the samples can be treated as aseries of curves, so that we can analyze it as function. Therefore, apply functional dataanalysis to quantitative investment and build up a trading system based on the price curverecognition of functional data analysis model, is the key of this paper.First, we establish a price curve recognition model based on functional data analysis. Useprincipal components analysis for functional data of price curve to explain the differenceeigenvalue among each curve. And then choose the first principal component function asinput for K-Means cluster analysis. After clustering, we found that20different centroidfunctions showed significant differences.Second, we built a future trading system, design the framework, including the selectionof evaluation indexes, the design of extrapolation testing and optimization design. And thenwe design the trading strategies based on the price curve recognition model of functional dataanalysis.Finally, we evaluate the result of extrapolation test. The report shows that the model has10%annualized returns and only2%trading days encounter more than1%losses in one day.But the maximum retracement up to8%, which means there exists some long term risks.