Cycle Forecasting in Financial Markets with LSTM and Discrete Fourier Transform

The study of cyclicality in capital markets is important for forecasting the prices of traded financial instruments and improving investment strategies. Cycles can be influenced by various factors – economic conditions, business activity, political events, and even market participants’ psychology – making their analysis challenging. This paper presents a study aimed at detecting cyclicality in time series using a hybrid model. Recurrent neural networks of the Long Short-Term Memory (LSTM) type are combined with Discrete Fourier Transform for frequency component analysis. The model is trained on simulated market data containing clearly defined sinusoidal cycles and stochastic noise to assess the LSTM network’s ability to capture the period, frequency, and amplitude of market fluctuations. Experiments were conducted with varying numbers of hidden neurons in the LSTM layer, and the results show that increasing architectural complexity improves amplitude prediction, while frequency and period are captured with high accuracy across all configurations.