LSTMs analyze shock propagation in economic systems using the LSTM multiplier response function, capturing nonlinear dynamics and current states more effectively than traditional VAR models, enhancing forecasting accuracy.
Introducing LSTM Multiplier Response Functions
Long-short term memory networks (LSTMs) offer an innovative approach for studying shock propagation in complex economic and financial systems. This working paper presents the LSTM multiplier response function, akin to the impulse response function in a linear VAR model. The LSTM approach provides distinct advantages, including the ability to capture nonlinear dynamics that often exist in complex systems, unlike linear VAR models.
Advantages of LSTM in Economic Analysis
LSTMs also consider the system’s current and past states, which is crucial because negative shocks tend to have greater impact during adverse conditions. Additionally, the LSTM approach allows for the direct application of shocks to relevant variables, removing the need to establish causality or orthogonalize the system. This simplifies the analysis and potentially leads to more accurate insights.
Comparing LSTM and VAR Models
The paper compares LSTM and VAR models within a multivariate economic context. By leveraging LSTM’s superior forecasting abilities, it demonstrates that the LSTM multiplier response function exhibits similar qualitative features to traditional VAR impulse responses. This finding underscores the potential of LSTM models in economic and financial applications, highlighting their practical utility and efficiency.
Source: Shock Propagation in LSTM Multivariate Time Series Systems – ASEAN+3 Macroeconomic Research Office
