Convolutional Neural Networks And Unconventional Data - Predicting The Stock Market Using Images

This presentation is about taking numerical data, transforming it into images and modeling it with convolutional neural networks. In other words, we are creating synthetic image-representation of quantitative data, and seeing if a CNN brings any additional modeling understanding, accuracy, a boost via ensembling, etc. CNN’s are hyper-contextually aware - they know what’s going on above, below, to the right and to the left, even what is going on in front and behind if you use image depth. Financial Technical Analysis is a visual art form that looks for patterns off graphical stock charts. Quantitative financial analysis relies on numerical data, programming and time-series events to evaluate opportunities, abnormalities, arbitrages, etc. These approaches seem to have little in common, yet, with the popularization and simplicity of today’s convolutional neural networks (CNNs), we are able to merge both and find new ways of understanding financial markets. At their core, all statistical models look at numbers. CNNs are no exception and its numbers start at the pixel level. But its level of awareness quickly extends beyond statistical time-series and leap into a complex level of multi-dimensional contextual awareness that gets closer to the inter-connectedness and complexity of the financial markets.