Intraday Volume Curve Prediction
41 Pages Posted: 9 Jan 2021
Date Written: March 13, 2015
We review some methodologies used to predict the intraday volume percentage curve, the intraday volumes as well as the closing auction volume. The methods can be very simple (average of historical curves), parametric (cubic function) or very sophisticated (linear model with real-time indicators). We use intraday data of CAC40 stocks from 01 April, 2014 to 31 December, 2014 to back-test these methods. We found that the best method to forecast the intraday percentage curve is the cubic method, justifying by volume-based metrics like Mean of Absolute Error (MAE) or Mean of Squared Error (MSE). However, when replacing the metric by the tracking error (TE), the non-linear model, fitting a regression tree with real time indicators, performs better than the others. It reduces the mean and median of TE 2%, compared with the historical median method. Concerning the prediction of raw volumes, the polynomial with correction outperforms the others in reduction the MAE 25% in mean and 31% in standard deviation compared to the historical method. The same method also provides the best forecast for closing volume. Evidence confirming the robustness of the improvement is also discussed and provided.
Keywords: intraday volume, auction volume, linear model, seasonality, cubic function, regression tree
JEL Classification: C13, C53,G12
Suggested Citation: Suggested Citation