Conditional Time Series Forecasting with Convolutional Neural Networks
By Anastasia Borovykh, Sander M. Bohte, Cornelis W. Oosterlee. 2017
591 citations recorded by the source corpus.
textabstractForecasting financial time series using past observations has been a significant topic of interest. While temporal relationships in the data exist, they are difficult to analyze and predict accurately due to the non-linear trends and noise present in the series. We propose to learn these dependencies by a convolutional neural network. In particular the focus is on multivariate time series forecasting. Effectively, we use multiple financial time series as input in the neural network, thus conditioning the forecast of a time series x(t) on both its own history as well as that of a second (or third) time series y(t). Training a model on multiple stock series allows the network to exploit the correlation structure between these series so that the network can learn the market dynamics in shorter sequences of data. We show that long-term temporal dependencies in and between financial time series can be learned by means of a deep convolutional neural network based on the WaveNet model [2]. The network makes use of dilated convolutions applied to multiple time series so that the receptive field of the network is wide enough to learn both short and long-term dependencies. The architecture includes batch normalization and uses a 1 × k convolution with parametrized skip connections from the input time series as well as the time series we condition on, in this way learning long-term interdependencies in an efficient manner [1]. This improves the forecast, while at the same time limiting the requirement for a long historical price series and reducing the noise. Knowing the strong performance of CNNs on classification problems we show that they can be applied successfully to forecasting financial time series, without the need of large samples of data. We compare the performance of the WaveNet model to a state-of-the-art fully convolutional network (FCN), and an autoregressive model popular in econometrics and show that our model is much better able to learn important dependencies in between financial time series resulting in a more robust and accurate forecast.
Subfields: Economics and Econometrics, Management Science and Operations Research, Model Architectures, Signal Processing, Time-Series & Forecasting
How researchers see this paper
The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.
In Economics and Econometrics
- Dhruv Madeka
- Laurent Callot
- Mu-Hwa Chen
- Rahim Barzegar
- Michael Bohlke-Schneider
- Jan Adamowski
- Bryan Lim
- David Salinas
In Management Science and Operations Research
- Bryan Lim
- Bernie Wang
- Jan Gasthaus
- Stefan Zohren
- Tim Januschowski
- Haixu Wu
- Mingsheng Long
- Syama Sundar Rangapuram
In Model Architectures
In Signal Processing
- Bryan Lim
- Haixu Wu
- Bernie Wang
- Stefan Zohren
- Jan Gasthaus
- Mingsheng Long
- Tim Januschowski
- Jianmin Wang
In Time-Series & Forecasting
- Bryan Lim
- Bernie Wang
- Jan Gasthaus
- Stefan Zohren
- Tim Januschowski
- Syama Sundar Rangapuram
- Haixu Wu
- Mingsheng Long
Citations in the corpus
- Time-series forecasting with deep learning: a survey (2020, methodology): “To adapt CNNs to time series datasets, researchers utilise multiple layers of causal convolutions [23, 24, 25] – i.e. convolutional filters designed to ensure only past information is used for forecasting.”
- Think Globally, Act Locally: A Deep Neural Network Approach to High-Dimensional Time Series Forecasting (2019, methodology): “TCN’s have also been shown to require apriori normalization [3] for time-series predictions.”
- A Multi-Horizon Quantile Recurrent Forecaster (2017, methodology): “Notably, Convolutional neural networks (CNN) with Seq2Seq structures have gained recent popularity, after the success of WaveNet (Van Den Oord et al, 2016) in the field of audio generation, and have also been studied under the topic of forecasting (Borovykh et al, 2017).”
- Autoregressive Convolutional Neural Networks for Asynchronous Time Series (2017, methodology): “More recent papers include (Sirignano, 2016) who used 4layer perceptrons in modeling price change distributions in Limit Order Books, and (Borovykh et al., 2017) who applied WaveNet architecture of (van den Oord et al.”
- MICN: Multi-scale Local and Global Context Modeling for Long-term Series Forecasting (2023, background): “It can integrate the local information of the sequence better and achieve competitive results in short and medium-term forecasting (Sen et al., 2019) (Borovykh et al., 2017).”
- Deep reinforcement learning for time series: playing idealized trading games (2018, methodology): “Borovykh et al. [19] proposed a method to forecast multivariate time series based on deep convolutional WaveNet architecture.”
- A CNN-RNN Framework for Crop Yield Prediction (2019, methodology): “Similar convolutional models have been widely used in various application domains and found to be effective in improving prediction accuracy (Ince et al., 2016; Borovykh et al., 2017; Kiranyaz et al., 2019).”
- Adaptive Graph Convolutional Recurrent Network for Traffic Forecasting (2020, background): “Some efforts employ temporal convolutional networks [21, 22, 23] to enable the model process very long sequence with fewer time.”
- SCINet: Time Series Modeling and Forecasting with Sample Convolution and Interaction (2021, methodology): “Another popular type of TSF model is the so-called temporal convolutional network [7, 4, 33, 39, 27], wherein convolutional filters are used to capture local temporal features (see Fig.”
- Autoformer: Decomposition Transformers with Auto-Correlation for Long-Term Series Forecasting (2021, background): “Also, many works based on temporal convolution networks (TCN) [40, 5, 4, 35] attempt to model the temporal causality with the causal convolution.”
- A Survey on Graph Neural Networks for Time Series: Forecasting, Classification, Imputation, and Anomaly Detection (2023, background): “The advent of deep learning technologies has led to the development of different neural networks based on convolutional neural networks (CNN) [24], [25], recurrent neural networks (RNN) [26], and transformers [27], which have shown significant advantages in modeling real-world time series data.”
- A decoder-only foundation model for time-series forecasting (2023, methodology): “(ii) Global univariate models like DeepAR (Salinas et al., 2020), Temporal Convolutions (Borovykh et al., 2017), N-BEATS (Oreshkin et al., 2019) and long-term forecasting models such as (Nie et al., 2022; Das et al., 2023) that are trained globally on many time-series but during inference they predi”
- Deep Learning for Time Series Forecasting: Tutorial and Literature Survey (2020, methodology): “[24] CNN Point 1 l1 , l2 regularizer RMSE, MASE, HITS index forecasting, exchange rate WaveNet [183] based model adjusted for time series forecasting”
- Multi-step Deep Autoregressive Forecasting with Latent States (2019, methodology): “Most existing works employ a Recurrent Neural Network (RNN) (Lai et al., 2018; Li et al., 2017; Yu et al., 2017; Flunkert et al., 2017) or a Temporal Convolutional Network (TCN) (Borovykh et al., 2017) to model an AR process.”
- Deep Learning for Time Series Forecasting: The Electric Load Case (2019, methodology): “In our work we focus on a convolutional architecture inspired by Wavenet [85], a fully probabilistic and autoregressive model used for generating raw audio wave-forms and extended to time-series prediction tasks [87].”
- Short-term water quality variable prediction using a hybrid CNN–LSTM deep learning model (2020, methodology): “f , whose size depends on the filter size and number of filters used in the final layer (Borovykh et al. 2017).”