Abstract:Purpose Because the dynamic characteristics of the stock market are nonlinear, it is unclear whether stock prices can be predicted. This paper aims to explore the predictability of the stock price index from a long-memory perspective. The authors propose hybrid models to predict the next-day closing price index and explore the policy effects behind stock prices. The paper aims to discuss the aforementioned ideas. Design/methodology/approach The authors found a long memory in the stock price index series using modified R/S and GPH tests, and propose an improved bi-directional gated recurrent units (BiGRU) hybrid network framework to predict the next-day stock price index. The proposed framework integrates (1) A de-noising module—Singular Spectrum Analysis (SSA) algorithm, (2) a predictive module—BiGRU model, and (3) an optimization module—Grid Search Cross-validation (GSCV) algorithm. Findings Three critical findings are long memory, fit effectiveness and model optimization. There is long memory (predictability) in the stock price index series. The proposed framework yields predictions of optimum fit. Data de-noising and parameter optimization can improve the model fit. Practical implications The empirical data are obtained from the financial data of listed companies in the Wind Financial Terminal. The model can accurately predict stock price index series, guide investors to make reasonable investment decisions, and provide a basis for establishing individual industry stock investment strategies. Social implications If the index series in the stock market exhibits long-memory characteristics, the policy implication is that fractal markets, even in the nonlinear case, allow for a corresponding distribution pattern in the value of portfolio assets. The risk of stock price volatility in various sectors has expanded due to the effects of the COVID-19 pandemic and the R-U conflict on the stock market. Predicting future trends by forecasting stock prices is critical for minimizing financial risk. The ability to mitigate the epidemic's impact and stop losses promptly is relevant to market regulators, companies and other relevant stakeholders. Originality/value Although long memory exists, the stock price index series can be predicted. However, price fluctuations are unstable and chaotic, and traditional mathematical and statistical methods cannot provide precise predictions. The network framework proposed in this paper has robust horizontal connections between units, strong memory capability and stronger generalization ability than traditional network structures. The authors demonstrate significant performance improvements of SSA-BiGRU-GSCV over comparison models on Chinese stocks.

A Long-Memory Model for Multiple Cycles with an Application to the US Stock Market

Extending the Fama-French Five-Facor Model with the Long Memory Factor Compared with Machine Learning

Modeling Long Cycles

Cyclical Long Memory: Decoupling, Modulation, and Modeling

Memory Effects, Multiple Time Scales and Local Stability in Langevin Models of the S&P500 Market Correlation

LONG MEMORY IN STOCK TRADING

Uncovering Long Memory in High Frequency UK Futures

Long Memory Process in Asset Returns with Multivariate GARCH Innovations

A Segmented Regime-Switching Model with Its Application to Stock Market Indices

Long memory in select stock returns using an alternative wavelet log-scale alignment approach

Emergence of long memory in stock volatility from a modified Mike-Farmer model

Modeling Long Memory in REITs

Isolating Financial Cycles Using the Fractional Cyclical Model in Selected Economies: 1970–2019

Detecting turning points of stock markets in China and the United States

Bayesian inference for long memory term structure models

Stock price index prediction based on SSA-BiGRU-GSCV model from the perspective of long memory

Endogenous cycles in heterogeneous agent models: a state-space approach

On Long Memory Origins and Forecast Horizons

Multivariate tempered stable model with long-range dependence and time-varying volatility

ESTIMATION OF A DYNAMIC MULTI- LEVEL FACTOR MODEL WITH POSSIBLE LONG-RANGE DEPENDENCE

Robust inference of panel data models with interactive fixed effects under long memory: A frequency domain approach