Short-Term Forecasting of Financial Time Series with Deep Neural Networks

In this work, a high-frequency strategy using Deep Neural Networks (DNNs) is presented. The input information to the DNN consists of: (i). Current time (hour and minute); (ii). the last n one-minute pseudo-returns, where n is the sliding window size parameter; (iii). the last n one-minute standard d...

Full description

Autores:: Arévalo Murillo, Andrés Ricardo

Tipo de recurso:

Fecha de publicación:: 2016

Institución:: Universidad Nacional de Colombia

Repositorio:: Universidad Nacional de Colombia

Idioma:: spa

Description
Summary:	In this work, a high-frequency strategy using Deep Neural Networks (DNNs) is presented. The input information to the DNN consists of: (i). Current time (hour and minute); (ii). the last n one-minute pseudo-returns, where n is the sliding window size parameter; (iii). the last n one-minute standard deviations of the price; (iv). The last n trend indicator, computed as the slope of the linear model fitted using the transaction prices inside a particular minute. The output DNN prediction is the next one-minute pseudo-return, this output is later transformed to obtain the next one-minute average price forecasting. The DNN predictions are used to build a high-frequency trading strategy that buys (sells) when the next predicted average price is above (below) the last closing price. This high-frequency trading strategy is only applicable to high liquidity stocks, because it requires to open and close positions in a time interval equal or less than one minute. For experimental testing, this work uses three datasets: (i). Apple stock (ticker: AAPL) from September to November of 2008. (ii). Apple stock (ticker: AAPL) from August of 2015 to August of 2016. (iii). Google stock (ticker: GOOG) from August of 2015 to August of 2016. Apple Inc. and Google Inc. are high liquidity stocks. The period of the first dataset covers the stock crash during the financial crisis of 2008. During this crash, the AAPL price suffered a dramatic fall from 172 to 98 dollars. This first dataset was chosen intentionally for demonstrate the performance of the proposed strategy under high volatility conditions. Whereas the second and third datasets were chosen in order to test the proposed strategy in normal market conditions. Multiple DNNs with different sliding window size parameter n and number of hidden layers L were trained. The best-performing-found DNN has a 65% of directional accuracy.

Short-Term Forecasting of Financial Time Series with Deep Neural Networks

Publicaciones similares