Research on Modeling and Forecasting High-Frequency Financial Data Based on Histogram Time Series

High-frequency financial data is a type of financial transaction sequence forecasted using time units of minutes or even seconds. This type of data is featured by its large scale, irregular time interval, multi-modality, and skewed distribution, among other features that make it very challenging to model and perform inference. Traditional high-frequency data modeling strategies often involve downsampling to convert high-frequency data into low-frequency series for convenient analysis. into low-frequency data for modeling, this usually results in a great loss of information. On the contrary, constructing models based on the empirical distribution of the observed data can effectively avoid this weakness. Therefore, based on the histogram time series, two new methods are proposed in this paper: a Histogram Autoregressive Model (HAR model) and a Histogram Threshold Autoregressive Model. The HTS (Histogram Time Series) demonstrates. With excellent information integration capabilities, these can also portray the accurate trend of phenomena over time, and thus, both regression models are highly flexible and effective to handle complex high-frequency data.