Preeti
In the electronic security trading system, limit orders are offered by potential buyers and sellers, and a trade will only be executed if the bid price is higher than the asked price. Trade duration prediction is important for traders executing a large number of trades. The autoregressive conditional duration (ACD) model is commonly used to model trade duration sequences. Various extensions to the ACD model have been proposed, including non-linear equations for mean duration and different error term distributions. In 2018, a new ZIACD model based on the zero-inflated negative binomial distribution was introduced to address excessive zero values in trade durations.
Recently, machine learning methods have gained attention in financial prediction tasks. Random forests, support vector regression, and deep neural networks have been applied successfully. A new hybrid model called DL-ZIACD was proposed in this study, combining a zero-inflated duration model with a deep learning network to forecast time-varying parameters of conditional duration distribution. The model was tested on constituent stocks of the Chinese Shenzhen Stock Exchange 100 Index and outperformed benchmark models in forecasting conditional duration distribution.
The study contributes by proposing a new hybrid model for conditional duration distribution, considering the behavioral differences of large traders and small individual traders. The model showed superior forecasting performance compared to benchmark models. The empirical results demonstrated the accuracy and robustness of the DL-ZIACD model in forecasting both central location and extreme quantiles of conditional duration distribution. The model can be applied beyond financial transaction duration analysis and has potential for other time series forecasting tasks.
Source link
Source link: https://www.frontiersin.org/journals/physics/articles/10.3389/fphy.2021.651528/full
GIPHY App Key not set. Please check settings