Forecasting Airtel Stock Prices Through Decomposition and Integration: A Novel VMD‐GARCH‐LSTM Framework

Source abstract

Stock price forecasting is complex due to the nonlinear and nonstationary nature of financial time series. This study proposes a hybrid variational mode decomposition (VMD)–generalized autoregressive conditional heteroskedasticity (GARCH)–long short‐term memory (LSTM) model to predict Airtel’s stock prices, integrating VMD, GARCH, and LSTM networks. VMD decomposes the stock price series into intrinsic mode functions (IMFs), enabling frequency‐specific modeling. High‐frequency IMFs, which exhibit volatility, are processed with GARCH to capture time‐varying volatility and are then fed into LSTM to model residual nonlinear dynamics. Low‐frequency IMFs, which reflect smoother trends, are directly modeled by LSTM. Final forecasts are aggregated via an additive ensemble. Initially evaluated on an 80:20 train–test split, the model underwent robustness checks with 70:30 and 90:10 splits, consistently outperforming benchmark models: transformer, GRU, LSTM, BiGRU, BiLSTM, VMD‐transformer, VMD‐GRU, VMD‐LSTM, VMD‐BiGRU, VMD‐BiLSTM, VMD‐GARCH‐transformer, VMD‐GARCH‐GRU, VMD‐GARCH‐BiGRU, and VMD‐GARCH‐BiLSTM. Diebold–Mariano tests confirmed statistical superiority across MSE, MAE, and MAPE loss functions, validating the model’s enhanced accuracy. These robustness checks demonstrate the model’s stability across varied data partitions, offering investors a reliable tool for risk mitigation. By effectively addressing volatility and nonlinearity, this hybrid framework advances financial forecasting, enhances decision‐making in dynamic financial stocks, and contributes to robust time‐series prediction methodologies.