On The Detection Of Breaking Points In High-frequency Financial Time Series: A LLSA Wavelet Perspective

Stock price changes are caused by the arrival of information. Therefore, theaccurately and timely access to new information is important to reveal the inherentmechanism of the stock price. In the time series analysis of stock price, in addition tolong-term trends and seasonality, there is another change which is also caused by theexternal events and will have a continuing impact on the trend of the time series, i.e.,breaking points, including jumps, steep slopes, etc. It often contains importantinformation but is viewed mistakenly as noise and then be ignored. Breaking points inthe high-frequency financial time series often contain important information, whichprotrudes the important role of accurate detection and analysis in making investmentdecisions. Statistical data mining methods (or models) are in urgent need of ade-noising algorithm to clean the data, to obtain reliable and significant results. Mostdata cleaning methods only focus on some of the known types of irregular behavior.For high-frequency financial data, the irregularities are complex. Therefore, findingan effective de-noising algorithm is the key to high-frequency financial data miningproceed.Previous studies had some limitations. They cannot keep a balance betweenensuring that the trend does not contain too much noise extraction and containingcertain breaking points. Conversely, if you want to detect breaking points, the costpaid into noise is unnecessary. This paper employs an improved wavelet analysismethod-locally linear scale approximation based on the MODWT, combined with thelinear and nonlinear filters characteristics of the high-frequency financial data formutation detection and reconstruction. The empirical results show that this methodcan effectively detect breaking points. And breaking points corresponds to somemajor economic events. The reconstructed time series fit the actual data better thanthe MODWT method and improve the prediction accuracy. Besides, the usability isalso test from the viewpoint of multi-time scale and economic significance is tested.