MODELING FUTURES TRADING VOLUME WITH NON-PARAMETRIC ADDITIVE MEM: EVIDENCE FROM CHINESE FUTURES MARKETS

What the paper says

Futures trading volume, as a direct indicator of market scale and liquidity, plays a critical role in reflecting market activity, price discovery efficiency, and risk characteristics. The multiplicative error model (MEM), designed to capture the dynamics of non-negative financial indicators, has been widely applied to variables such as trading volume, trading value, and volatility. Existing research has primarily focused on parametric MEM, while some studies have explored non-parametric extensions. Although non-parametric MEM offers greater flexibility and has been shown to better capture market volatility, it suffers from the “curse of dimensionality,” limiting its effectiveness in models with multiple lagged variables. To address this issue, this paper introduces the framework of non-parametric additive modeling into MEM and develops non-parametric additive MEM, estimated via penalized splines. Using Monte Carlo simulations, we compare the performance of parametric MEM, non-parametric MEM, and non-parametric additive MEM with different lag orders. The results show that estimation errors of non-parametric MEM increase markedly with higher lag orders, whereas non-parametric additive MEM maintains superior accuracy in multidimensional settings. Finally, an empirical analysis to futures trading volume in three Chinese futures markets demonstrates that non-parametric additive MEM(2, 2) achieves the highest estimation accuracy among the competing models.