[Paper Review] Recursive deep learning framework for forecasting the decadal world economic outlook

Tianyi Wang, Rodney Beard+2 more|arXiv (Cornell University)|Jan 25, 2023

Market Dynamics and VolatilityEconomics, Econometrics and Finance3 citations

TL;DR

This paper proposes a recursive deep learning framework using ED-LSTM to forecast decadal GDP growth rates across 13 countries from 1980 to 2019 using Penn World Table data. The ED-LSTM model outperformed ARIMA, VAR, LSTM, BD-LSTM, and CNN, predicting a global economic slowdown in the first five years of the next decade, with only China, France, and India expected to maintain stable or increasing growth.

ABSTRACT

The gross domestic product (GDP) is the most widely used indicator in macroeconomics and the main tool for measuring a country's economic output. Due to the diversity and complexity of the world economy, a wide range of models have been used, but there are challenges in making decadal GDP forecasts given unexpected changes such as emergence of catastrophic world events including pandemics and wars. Deep learning models are well suited for modelling temporal sequences and time series forecasting. In this paper, we develop a deep learning framework to forecast the GDP growth rate of the world economy over a decade. We use the Penn World Table as the data source featuring 13 countries prior to the COVID-19 pandemic, such as Australia, China, India, and the United States. We present a recursive deep learning framework to predict the GDP growth rate in the next ten years. We test prominent deep learning models and compare their results with traditional econometric models for selected developed and developing countries. Our decadal forecasts reveal that that most of the developed countries would experience economic growth slowdown, stagnation and even recession within five years (2020-2024). Furthermore, our model forecasts show that only China, France, and India would experience stable GDP growth.

Motivation & Objective

• To develop a robust deep learning framework for long-term GDP growth forecasting across multiple countries.
• To address the limitations of traditional linear time series models (e.g., ARIMA, VAR) in capturing non-linear economic dynamics and sudden shocks.
• To evaluate the performance of diverse deep learning architectures—LSTM, BD-LSTM, ED-LSTM, and CNN—against classical econometric models.
• To identify which deep learning model best captures complex temporal dependencies in macroeconomic time series for decadal forecasting.
• To project global economic trends over the next decade, particularly under conditions of structural change and external shocks like pandemics and wars.

Proposed method

• Utilized the Penn World Table dataset covering 13 countries from 1980 to 2019 for GDP and related macroeconomic indicators.
• Implemented a recursive forecasting strategy where predictions for each year are fed back as inputs for the next year’s forecast.
• Applied data shuffling during training to prevent overfitting and improve generalization across countries and time periods.
• Evaluated multiple deep learning models: LSTM, bidirectional LSTM (BD-LSTM), encoder-decoder LSTM (ED-LSTM), and Convolutional Neural Networks (CNN).
• Compared model performance against traditional time series models: ARIMA and Vector Autoregression (VAR).
• Selected ED-LSTM as the optimal model based on lowest mean squared error (MSE) and best predictive accuracy across test sets.

Figure 1 : The process of using VAR.

Experimental results

Research questions

• RQ1 Which deep learning architecture—LSTM, BD-LSTM, ED-LSTM, or CNN—performs best in forecasting decadal GDP growth rates?
• RQ2 How does the recursive deep learning framework improve long-term forecasting accuracy compared to direct or non-recursive approaches?
• RQ3 To what extent can deep learning models capture structural economic shifts and external shocks (e.g., pandemics, wars) in long-horizon GDP forecasts?
• RQ4 How do the forecasted GDP growth trends compare across developed and developing economies over the next decade?
• RQ5 Can the ED-LSTM model reliably predict sustained growth in countries like China, India, and France amid global economic turbulence?

Key findings

• The ED-LSTM model achieved the lowest mean squared error (MSE) and outperformed ARIMA, VAR, LSTM, BD-LSTM, and CNN in decadal GDP growth forecasting.
• The recursive strategy enabled consistent long-term predictions by iteratively feeding predicted values back into the model for subsequent time steps.
• The model predicted a global economic slowdown or stagnation in the first five years of the next decade, with many countries facing negative growth or recession.
• Only China, France, and India were forecasted to maintain stable or increasing GDP growth rates over the full decade, with China projected to sustain high growth despite economic scale.
• The model’s predictions align with external reports forecasting India’s GDP to reach $11 trillion by 2032 and China’s shift toward high-quality growth post-2020.
• Limitations include the exclusion of post-2020 data (e.g., pandemic and war impacts), and challenges in modeling decelerating growth as economies mature.

Figure 2 : RNN structure showing information from input $x$ to output $y$ via the state and hidden layers $h$ .