A Predictive Framework for Financial Crashes Using Advanced Time Series Techniques

The research explores how financial markets, specifically the NASDAQ and S&P 500 indices, react under pressure and seeks to predict market crashes using advanced time series techniques. Financial market crashes, such as those in 2008 and during the COVID-19 pandemic, have significantly impacted the global economy, making accurate crash prediction essential. The study employs ARIMA and GARCH models to forecast market volatility, integrating these models with dynamic GARCH extensions and wavelet-based time series decomposition to capture non-linear market behavior. Data from January 2021 to August 2024 was collected, with 921 observations for NASDAQ and 787 for S&P 500. The stationarity of the time series was confirmed through ADF and PP tests, supporting the use of ARIMA and GARCH models. The GARCH model detected volatility clustering, where periods of high volatility were followed by further increases, particularly in NASDAQ, with a GARCH effect of 0.912741. For S&P 500, the GARCH effect was lower at 0.6785330, indicating less volatility persistence. This study also employs a Recurrent Neural Network (RNN) model to showcase thr capability of deep learning models to predict financial crashes. The RNN model with an MAPE of 5.18%, performs very well. These findings highlight the effectiveness of the predictive framework in forecasting market crashes, offering valuable insights for investors and policymakers.