Using Self-Organizing Maps to Adjust Intra-Day Seasonality

Abstract

The existence of an intra-day seasonality component within financial market variables (volatility, volume, activity,. . .), has been highlighted in many previous works. To adjust raw data from their cyclical component, many studies start by implementing the intra-daily average observations model (IAOM) and/or some smoothing techniques (e.g. the kernel method) in order to remove the day of the week effect. When seasonality involves only a deterministic component, IAOM method succeed in estimating periodicity almost without estimation error. However, when seasonality contains both deterministic and stochastic components (e.g. closed days), we show that either the IAOM or the kernel method fail to capture it. We introduce the use of the self-organizing maps (SOM) as a solution. SOM are based on neural network learning and nonlinear projections. Their flexibility allows capturing seasonality even in the presence of stochastic cycles.