Research On The Duration Of Gold Futures Market Based On ACD Model

The duration(or trading interval)is an important variable to understand the microstructure and market behavior.To some extent,it can reflect the decision-making process of market participants,and helps to understand the dynamic change of the transaction process.Based on the theory of market microstructure,this paper focuses on the commodity futures market,and analyzes the duration of price,volume and open interest of the gold futures market.We choose five minutes gold futures trading data of Shanghai gold futures market as the research object.Based on the linear ACD model,the exponential distribution,Weibull distribution,Gamma distribution and Frechet distribution respectively as the error distribution,using a variety of quantitative evaluation criteria to compare several ACD model estimation results.The results show that the lagged price,trading volume and open interest duration model constructed by high frequency data,WACD model is relatively better performance,that is to say the Weibull distribution is more suitable for the description of China's gold futures market of high-frequency financial time sequence data.Through the analysis of the characteristics of gold futures data days and micro influence research,we found that the transaction time interval of China's gold futures market has obvious Intraday Effect,for price duration,it shows "N" mode in the forenoon,and "U" in the afternoon,for trading volume and open interest duration,they show "dual N" mode both in morning and afternoon.The duration of the gold futures market has apparent sustained and agglomeration characteristics.Based on the WACD model,the influence factors of each renewal were modeled by adding micro factors respectively.It was found that most of the micro factors had significant explanatory effects on the corresponding duration.It is well known that most of the parameters are estimated by maximum likelihood estimation or pseudo maximum likelihood estimation in the ACD model.Due to the thick tailed nature of gold futures market data(verified by Hill estimation chart),(pseudo)maximum likelihood estimation is not robust enough.After numerical simulation,it is found that when the error obeys the heavy tailed distribution or the residual sequence has outliers,the ACD model of the least absolute deviation estimation is more robust than the maximum likelihood estimation.In view of this,we use the least absolute deviation estimation method to model ACD,as a correction of the previous modeling results.