A CNN–BiLSTM Architecture for Macroeconomic Time Series Forecasting
In forecasting socio-economic processes, it is essential to have tools that are highly performing, with results as close to reality as possible. Forecasting plays an important role in shaping the decisions of governments and central banks about macroeconomic planning, and it is an essential analytic...
| Main Author: | |
|---|---|
| Format: | Article |
| Language: | English |
| Published: |
MDPI AG
2023-06-01
|
| Series: | Engineering Proceedings |
| Subjects: | |
| Online Access: | https://www.mdpi.com/2673-4591/39/1/33 |
| _version_ | 1797580355982589952 |
|---|---|
| author | Alessio Staffini |
| author_facet | Alessio Staffini |
| author_sort | Alessio Staffini |
| collection | DOAJ |
| description | In forecasting socio-economic processes, it is essential to have tools that are highly performing, with results as close to reality as possible. Forecasting plays an important role in shaping the decisions of governments and central banks about macroeconomic planning, and it is an essential analytical tool in defining economic strategies of countries. The most common forecasting methods used in the analysis of macroeconomic processes are based on extrapolation, i.e., extending the trend observed in the past (and present) to the future. However, the presence of non-linearity in the socio-economic systems under uncertainty, as well as the partial observability of the processes, has contributed to make researchers and practitioners consider other methodologies, too. In this paper, we analyze 18 time series of macroeconomic variables of the United States of America. We compare the benchmark results obtained with “classic” forecasting techniques with those obtained with our proposed architecture. The model we construct can be defined as “hybrid” since it combines a Convolutional Neural Network (CNN) with a Bidirectional Long Short-Term Memory Network (BiLSTM) backend. We show that, for what concerns minimizing the forecast error, our model competes with and often improves the results obtained with the benchmark techniques. The goal of this work is to highlight that, due to the recent advances in computing power, new techniques can be added to the set of tools available to a policymaker for forecasting macroeconomic data. |
| first_indexed | 2024-03-10T22:48:46Z |
| format | Article |
| id | doaj.art-f34b2226060549d48bc5c5266c9fb57d |
| institution | Directory Open Access Journal |
| issn | 2673-4591 |
| language | English |
| last_indexed | 2024-03-10T22:48:46Z |
| publishDate | 2023-06-01 |
| publisher | MDPI AG |
| record_format | Article |
| series | Engineering Proceedings |
| spelling | doaj.art-f34b2226060549d48bc5c5266c9fb57d2023-11-19T10:30:39ZengMDPI AGEngineering Proceedings2673-45912023-06-013913310.3390/engproc2023039033A CNN–BiLSTM Architecture for Macroeconomic Time Series ForecastingAlessio Staffini0Department of Economics and Finance, Catholic University of Milan, 20123 Milan, ItalyIn forecasting socio-economic processes, it is essential to have tools that are highly performing, with results as close to reality as possible. Forecasting plays an important role in shaping the decisions of governments and central banks about macroeconomic planning, and it is an essential analytical tool in defining economic strategies of countries. The most common forecasting methods used in the analysis of macroeconomic processes are based on extrapolation, i.e., extending the trend observed in the past (and present) to the future. However, the presence of non-linearity in the socio-economic systems under uncertainty, as well as the partial observability of the processes, has contributed to make researchers and practitioners consider other methodologies, too. In this paper, we analyze 18 time series of macroeconomic variables of the United States of America. We compare the benchmark results obtained with “classic” forecasting techniques with those obtained with our proposed architecture. The model we construct can be defined as “hybrid” since it combines a Convolutional Neural Network (CNN) with a Bidirectional Long Short-Term Memory Network (BiLSTM) backend. We show that, for what concerns minimizing the forecast error, our model competes with and often improves the results obtained with the benchmark techniques. The goal of this work is to highlight that, due to the recent advances in computing power, new techniques can be added to the set of tools available to a policymaker for forecasting macroeconomic data.https://www.mdpi.com/2673-4591/39/1/33time series forecastingeconomic forecastingmacroeconometric forecastingdeep learningCNN-BiLSTM |
| spellingShingle | Alessio Staffini A CNN–BiLSTM Architecture for Macroeconomic Time Series Forecasting Engineering Proceedings time series forecasting economic forecasting macroeconometric forecasting deep learning CNN-BiLSTM |
| title | A CNN–BiLSTM Architecture for Macroeconomic Time Series Forecasting |
| title_full | A CNN–BiLSTM Architecture for Macroeconomic Time Series Forecasting |
| title_fullStr | A CNN–BiLSTM Architecture for Macroeconomic Time Series Forecasting |
| title_full_unstemmed | A CNN–BiLSTM Architecture for Macroeconomic Time Series Forecasting |
| title_short | A CNN–BiLSTM Architecture for Macroeconomic Time Series Forecasting |
| title_sort | cnn bilstm architecture for macroeconomic time series forecasting |
| topic | time series forecasting economic forecasting macroeconometric forecasting deep learning CNN-BiLSTM |
| url | https://www.mdpi.com/2673-4591/39/1/33 |
| work_keys_str_mv | AT alessiostaffini acnnbilstmarchitectureformacroeconomictimeseriesforecasting AT alessiostaffini cnnbilstmarchitectureformacroeconomictimeseriesforecasting |