Volume Prediction With Neural Networks

Abstract

Changes in intraday trading volume are integral to any algorithmic trading strategy. Accordingly, forecasting the change in trading volume is paramount to better understanding the financial markets. This paper introduces a new method to forecast the log change in trading volume, leveraging the power of Long Short Term Memory (LSTM) networks in conjunction with Support Vector Regression (SVR) and Autoregressive (AR) models. We show that LSTM contributes to a more accurate forecast, particularly when constructed as part of a hybrid model with AR. The algorithm is extended to include data about the time of day, helping the model associate the log change in trading volume with the current hour, which yields the best performance of all trials.

Long-term, Short-term and Sudden Event: Trading Volume Movement Prediction with Graph-based Multi-view Modeling

“…Because of the noisy fluctuations existed in the trading data, it is a desirable ability to incorporate longterm information to alleviate the short-term noise. In contrast with the previous work [Libman et al, 2019], which views the price and volume data as a whole and feeds them into a neural network to get a mixed representation, we argue that there is a heterogeneity gap between price data and volume data, therefore we separately process price data and volume data as price-graph and volume-graph. Each layer of our short-term fluctuation module consists of two parts, aggregation step and update step.…”

“…Trading volume movement prediction aims to predict the volume in a certain period of time based on stock market information, which is crucial to a variety of financial applications, e.g., stock market anomaly detection, risk management and algorithmic trading [Brownlees et al, 2010;Libman et al, 2019]. More importantly, when investors try to buy/sell large quantities of stocks, the order itself will instantaneously drive the stock price to the undesirable direction (i.e., higher price for a buy order and lower price for a sell order) and thus the total cost for the execution will be very expensive [Ye et al, 2014].…”

“…Previous works on volume prediction focus on predicting the volume based on short-term transaction data. Both classic machine learning models [Liu and Lai, 2017;Roman, 2018] and deep learning based models [Libman et al, 2019] have been applied. However, even though more sophisticated models like LSTM [Hochreiter and Schmidhuber, 1997] have been used to model the temporal sequence, almost all the previous work considers the transaction sequence in isolation.…”

“…Conventional methods tend to ignore long-term information [Brownlees et al, 2010;Liu and Lai, 2017;Roman, 2018;Libman et al, 2019], and experiments show that simply putting together information of different views cannot explicitly exploit the connection among them. In this paper, we propose to jointly integrate long-term, short-term and sudden event information into a graph via Correlation-powered Graph-based Multi-view (CGM) modeling method.…”

Forecasting Quoted Depth With the Limit Order Book

“…We utilize Deep Feed Forward Neural Networks for our forecasting algorithm. Deep learning has been steadily growing in popularity, especially in academic studies, which have shown successful applications of deep learning algorithms in a variety of settings (see for example, Libman et al, 2019 ). Despite this, papers researching applications of deep learning methods to the financial markets are often recent and few in number.…”