Start Date

16-8-2018 12:00 AM

Description

The forecasting of time series is a common problem in different domains. Especially the financial sector that relies heavily on algorithmic trading, employs intelligent systems for this purpose. With technological advances in the domain of intelligent systems those forecasts tend to become more and more precise. However, using such black box methods often lacks the understanding, whether the underlying time series characteristics like stationarity or chaos are caught by the corresponding algorithm and what particular external factors like dataset size influence the outcome. In addition to that the knowledge base is lacking a systematic overview of available algorithms and their application on different time series. We aim to contribute to the knowledge base by (1) conducting a literature review and descriptive analysis revealing dependencies between characteristics and algorithm use and (2) evaluating the impact of influence factors like dataset size and prediction window on the performance of Deep Learning Systems (DLS) in the context of crude oil price prediction, considering two common tasks: a trend and an exact value prediction problem.

Share

COinS
 
Aug 16th, 12:00 AM

Deep Intelligent Systems for Time Series Prediction: Champion or Lame Duck? - Evidence from Crude Oil Price Prediction

The forecasting of time series is a common problem in different domains. Especially the financial sector that relies heavily on algorithmic trading, employs intelligent systems for this purpose. With technological advances in the domain of intelligent systems those forecasts tend to become more and more precise. However, using such black box methods often lacks the understanding, whether the underlying time series characteristics like stationarity or chaos are caught by the corresponding algorithm and what particular external factors like dataset size influence the outcome. In addition to that the knowledge base is lacking a systematic overview of available algorithms and their application on different time series. We aim to contribute to the knowledge base by (1) conducting a literature review and descriptive analysis revealing dependencies between characteristics and algorithm use and (2) evaluating the impact of influence factors like dataset size and prediction window on the performance of Deep Learning Systems (DLS) in the context of crude oil price prediction, considering two common tasks: a trend and an exact value prediction problem.