STBM+: Advanced Stochastic Trading Behavior Model for Financial Markets using Residual Blocks or Transformers
This study proposes a new model to reverse engineer and predict traders’ behavior for the financial market. This trial is essential to build a more reliable simulation because the reliability of models is a fundamental issue in the increasing use of simulations. Thus, we tried to build a behavior mo...
Gespeichert in:
Veröffentlicht in: | New generation computing 2022, Vol.40 (1), p.7-24 |
---|---|
Hauptverfasser: | , , |
Format: | Artikel |
Sprache: | eng |
Schlagworte: | |
Online-Zugang: | Volltext |
Tags: |
Tag hinzufügen
Keine Tags, Fügen Sie den ersten Tag hinzu!
|
Zusammenfassung: | This study proposes a new model to reverse engineer and predict traders’ behavior for the financial market. This trial is essential to build a more reliable simulation because the reliability of models is a fundamental issue in the increasing use of simulations. Thus, we tried to build a behavior model of financial traders through the traders’ future action predicting using the actual order data. This study focused on one category of traders employing high-frequency market-making (HFT-MM) trading in financial markets. In our experiments, we build models for predicting the next actions of each trader and evaluate how correctly these models successfully predict trades’ future actions in the next one minutes. Although the task is the same as previous work, this study newly used an architecture based on the transformer and residual block, and a loss function based on the Kullback-Leibler divergence (KLD). In addition, we established a new evaluation metric. Consequently, our new models, both transformer-based and residual-block-based models, outperformed the previous model based on LSTM in terms of both old and new evaluation metrics. These results suggested that transformer and residual block are effective in capturing traders’ behaviors. In addition, the KLD-based new loss function also showed better results than the previous MSE-based loss function. We assumed it is because the KLD-based loss function has a better fitting to this task due to its mathematical form. |
---|---|
ISSN: | 0288-3635 1882-7055 |
DOI: | 10.1007/s00354-021-00145-z |