Futures price prediction method combining time delay analysis and hybrid model
By combining the Prophet-LSTM hybrid model with time delay analysis, the problems of nonlinearity and multi-factor influence in short-term gold futures price forecasting are solved, achieving more accurate short-term price forecasting.
Patent Information
- Application Number
- CN202610204669.5
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2026-02-12
- Publication Date
- 2026-05-01
AI Technical Summary
Existing technologies struggle to effectively handle the nonlinear characteristics and unforeseen events in financial market data for short-term gold futures price forecasting. Traditional models face challenges in terms of data requirements, computational resources, and interpretability, and they neglect the various factors influencing gold prices and their interactions.
A hybrid Prophet-LSTM model is adopted, which combines time lag analysis to group gold futures price data with multiple predictive factors. The Prophet model captures linear and periodic trends, while the LSTM model captures nonlinear trends. Accurate prediction is achieved by merging the predicted values.
It improves the accuracy of short-term gold futures price forecasts, better captures data features, and achieves accurate monthly forecasts, outperforming traditional algorithms and other allocation methods.
Smart Images
Figure CN121961730A_ABST
Abstract
Description
A Futures Price Forecasting Method Combining Time Delay Analysis and Hybrid Models Technical Field
[0001] This invention belongs to the field of gold futures price prediction technology, and in particular relates to a futures price prediction method that combines time lag analysis and hybrid models. Background Technology
[0002] Short-term forecasts of gold futures prices can help investors identify market trends and potential trading opportunities, enabling them to make more informed investment decisions. For short-term investors, accurately grasping short-term price fluctuations is crucial. By predicting potential short-term price movements, investors can implement specific risk control strategies, such as setting stop-loss levels, to reduce potential losses. Furthermore, short-term price forecasts help identify differences in gold prices across different markets, providing a basis for arbitrage trading. Short-term price fluctuations often reflect the impact of market sentiment and breaking news events; analyzing these factors helps in understanding market dynamics and investor behavior. However, the sheer volume, rapid changes, and complex influencing factors of financial market data mean that traditional statistical techniques are less effective at describing unstable or non-linear time series. This makes accurate price forecasting in financial markets difficult and imposes significant limitations in the analytical process, resulting in limited practical applicability.
[0003] Gold futures price forecasting is a complex and multidimensional research field, with scholars employing various methods to attempt to capture its dynamics and uncertainties. Traditionally, this field relies on statistical and econometric tools, such as time series analysis models, which learn from historical data and predict future trends. Furthermore, to address the common volatility clustering in financial data, autoregressive conditional heteroscedasticity models have been developed, specifically designed to model the volatility of financial time series. With advancements in computing power and data science, deep learning methods have begun to play a role in gold futures price forecasting in recent years. While statistical model-based financial data forecasting methods have achieved some success in historical data analysis and short-term forecasting, they have significant limitations in handling the nonlinear characteristics of financial market data, sudden events, and long-term forecasts. Neural network-based models have shown great potential in handling complex nonlinear financial data forecasting, but challenges remain regarding data requirements, computational resources, interpretability, and parameter tuning. Therefore, in practical applications, it is often necessary to combine neural network models with traditional statistical models, leveraging their respective advantages to achieve more accurate and reliable financial data forecasting. However, these models typically focus only on the characteristics of the data itself, ignoring the multiple factors that influence gold prices, including the independent effects of different factors and the combined effect of their interactions. Summary of the Invention
[0004] To address the aforementioned technical issues, this invention proposes a futures price forecasting method that combines time-delay analysis and a hybrid model. This method fully utilizes the time delay of data and the Prophet-LSTM hybrid model to improve the accuracy of short-term futures price forecasts.
[0005] To achieve the above objectives, this invention provides a futures price prediction method combining time-delay analysis and a hybrid model, comprising:
[0006] Acquire gold futures price datasets and forecast factor time series data, perform time lag analysis, and obtain forecast factors with optimal lag values less than the forecast length and forecast factors with optimal lag values greater than the forecast length.
[0007] By inputting the predictor with the best lag value less than the prediction length and the corresponding gold futures price into the Prophet model in the hybrid model, the fitted value, predicted value and residual value of the gold price are obtained.
[0008] By inputting the predictor with the best lag value greater than the prediction length and the residual value of the gold price into the LSTM model in the hybrid model, the residual prediction value of the gold price is obtained.
[0009] The predictions from the Prophet model and the residual predictions from the LSTM model are combined to obtain the prediction results for gold futures prices.
[0010] Optionally, obtaining predictors with optimal lag values less than the prediction length and predictors with optimal lag values greater than the prediction length includes:
[0011] Obtain a dataset of gold futures prices and predictive factors, analyze the correlation between gold futures prices and predictive factors, and determine the degree of influence of different predictive factors on gold futures prices.
[0012] Perform optimal lag analysis on each predictor to obtain the optimal lag value;
[0013] Analyze the relationship between the optimal lag value and the prediction length of each predictor to obtain predictors with optimal lag values less than the prediction length and predictors with optimal lag values greater than the prediction length.
[0014] Optionally, obtaining the fitted, predicted, and residual values for the gold price includes:
[0015] The predictor with the best lag value less than the prediction length and the corresponding gold futures price are input into the Prophet model in the hybrid model to predict the gold price.
[0016] Based on the Prophet model analysis, predictors with optimal lag values less than the prediction length and their corresponding gold futures prices are analyzed, and fitted and predicted values of the corresponding gold prices are generated.
[0017] The residual value of the gold price is obtained by calculating the actual value of the gold price and the fitted value.
[0018] Optionally, before obtaining the optimal lag value, it is necessary to analyze the correlation between the time series of gold futures prices and the time series of the predictor at different time lags to identify the potential relationship between the two series. The calculation is as follows:
[0019] ,
[0020] in, This represents the mean. and They are and The mean, and They are and standard deviation It is a time lag value.
[0021] Optionally, modeling gold prices includes:
[0022] The sliding window method was adopted, with the window size set to a preset size and the step size set to a preset length. The Prophet model moved across the entire dataset, moving a preset number of data points each time. Each window contained a preset number of data points for training and prediction. The performance of the Prophet model was observed in different time periods.
[0023] Optionally, the hybrid model is a Prophet-LSTM hybrid model.
[0024] Optionally, the predictive factors include other precious metal prices, crude oil futures prices, and macroeconomic indicators.
[0025] Optionally, the residual value of the gold price is the difference between the actual gold price and the fitted value obtained by training the Prophet model.
[0026] Technical advantages of this invention: This invention discloses a futures price prediction method combining time lag analysis and a hybrid model. The Prophet model is a seasonal decomposition method based on time series, capable of handling time series data with strong seasonal patterns. By combining the Prophet model with the LSTM model, and analyzing and integrating the time lags of various economic indicators and financial variables, the data is input into the Prophet-LSTM model according to different time lags, achieving accurate capture of short-term fluctuations in gold futures prices. The Prophet-LSTM hybrid model not only utilizes the advantages of the Prophet model in processing time series data, but also fully leverages the capabilities of the LSTM model in capturing long-term dependencies and handling nonlinear patterns, while making the most reasonable allocation of different predictive factors. This invention fully considers the time lag analysis of various price and macroeconomic environmental factor sequences that affect the trend of gold futures data. Compared with methods that do not consider predictive factors or other methods that incorporate predictive factors, Prophet, which considers immediate factors, and LSTM, which considers long-term factors, have the highest short-term prediction accuracy. Compared with traditional algorithms and other allocation methods, it can better capture data features on a monthly scale, achieving more accurate short-term data prediction. Attached Figure Description
[0027] The accompanying drawings, which form part of this application, are used to provide a further understanding of this application. The illustrative embodiments and descriptions of this application are used to explain this application and do not constitute an undue limitation of this application. In the drawings:
[0028] Figure 1 is a flowchart illustrating a futures price prediction method combining time delay analysis and a hybrid model according to an embodiment of the present invention.
[0029] Figure 2 is a comparison of the results between the Prophet-LSTM hybrid model with time delay, the Prophet model, and the true values in the embodiments of the present invention. Detailed Implementation
[0030] It should be noted that, unless otherwise specified, the embodiments and features described in this application can be combined with each other. This application will now be described in detail with reference to the accompanying drawings and embodiments.
[0031] It should be noted that the steps shown in the flowchart in the accompanying drawings can be executed in a computer system such as a set of computer-executable instructions, and although a logical order is shown in the flowchart, in some cases the steps shown or described may be executed in a different order than that shown here.
[0032] This invention considers that short-term forecasting focuses more on immediate factors and market dynamics. Therefore, it constructs an innovative hybrid forecasting model for gold futures prices by optimizing short-term forecasting through multiple factors. Time-lag analysis is performed on various price and macroeconomic environmental factor sequences affecting gold futures data trends. A hybrid futures price forecasting model combining time lags, the Prophet-LSTM hybrid model, is proposed to solve the problem of short-term forecasting of gold futures data. First, correlation and maximum time lag analysis are performed on each forecasting factor, and the forecasting factors are divided into immediate-influence factors and long-term-influence factors according to their time lags. Then, the immediate-influence factors are used to assist the Prophet model in capturing the linear and cyclical trends of gold futures, while the long-term-influence factors are used to assist the LSTM model in capturing the nonlinear trends of the gold futures forecasting residuals. Finally, the forecasting results using different categories of forecasting factors and different single and hybrid models are compared and analyzed based on commonly used indicators of forecasting algorithms.
[0033] As shown in Figure 1, this embodiment provides a futures price prediction method that combines time-delay analysis and a hybrid model, including:
[0034] Acquire gold futures price datasets and forecast factor time series data, perform time lag analysis, and obtain forecast factors with optimal lag values less than the forecast length and forecast factors with optimal lag values greater than the forecast length.
[0035] By inputting the predictor with the best lag value less than the prediction length and the corresponding gold futures price into the Prophet model in the hybrid model, the fitted value, predicted value and residual value of the gold price are obtained.
[0036] By inputting the predictor with the best lag value greater than the prediction length and the residual value of the gold price into the LSTM model in the hybrid model, the residual prediction value of the gold price is obtained.
[0037] The predictions from the Prophet model and the residual predictions from the LSTM model are combined to obtain the prediction results for gold futures prices.
[0038] Specifically, the predictive factor refers to the main factors that may affect the fluctuation of gold futures prices, as considered in this invention, such as opening price, highest price, lowest price, silver futures, gold prices of the Gold Miners ETF, and the US dollar index; the gold futures price is a time series, and the predictive factor time series data is also a time series.
[0039] Furthermore, obtaining predictors with optimal lag values less than the prediction length and predictors with optimal lag values greater than the prediction length includes:
[0040] Obtain a dataset of gold futures prices and predictive factors, analyze the correlation between gold futures prices and predictive factors, and determine the degree of influence of different predictive factors on gold futures prices.
[0041] Perform optimal lag analysis on each predictor to obtain the optimal lag value;
[0042] The relationship between the optimal lag value and the prediction length of each predictor is analyzed to obtain predictors with optimal lag values less than the prediction length and predictors with optimal lag values greater than the prediction length. The optimal lag value is obtained by calculating the Pearson correlation coefficient, and then the predictors are divided into two groups according to the optimal lag value. In this embodiment, the predictors with optimal lag values less than the prediction length are set as group X1, and the predictors with optimal lag values greater than the prediction length are set as group X2.
[0043] Furthermore, obtaining the fitted value, predicted value, and residual value of the gold price includes:
[0044] The predictor with the best lag value less than the prediction length and the corresponding gold futures price are input into the Prophet model in the hybrid model to predict the gold price.
[0045] Based on the Prophet model analysis, predictors with optimal lag values less than the prediction length and their corresponding gold futures prices are analyzed, and fitted and predicted values of the corresponding gold prices are generated.
[0046] The residual value of the gold price is obtained by calculating the difference between the actual gold price and the fitted value obtained from model training.
[0047] Specifically, a Prophet-LSTM hybrid model was first established, and then used to predict gold futures prices. When predicting gold futures prices, the impact of other predictive factors was considered. First, the Pearson correlation coefficient was calculated for these predictive factors, dividing them into two groups: a short-term predictive factor group, which was input into the Prophet model (outputting the predicted gold price), and a long-term predictive factor group, which was input into the LSTM model (outputting the predicted gold price residuals). Finally, the Prophet predictions and the LSTM residual predictions were summed to obtain the final result.
[0048] Furthermore, before obtaining the optimal lag value, it is necessary to analyze the correlation between the time series of gold futures prices and the time series of the predictor at different time lags to identify the potential relationship between the two series. The calculation is as follows:
[0049] ,
[0050] in, This represents the mean. and They are and The mean, and They are and standard deviation It is a time lag value.
[0051] Furthermore, modeling gold prices includes:
[0052] The sliding window method was adopted, with the window size set to a preset size and the step size set to a preset length. The Prophet model moved across the entire dataset, moving a preset number of data points each time. Each window contained a preset number of data points for training and prediction. The performance of the Prophet model was observed in different time periods.
[0053] Specifically, the sliding window method is used in the modeling process. The window size is set to 600 and the step size is 30. The model will move across the entire dataset, moving 30 data points each time. Each window contains 600 data points for training and prediction, so that the model's performance can be observed in detail over different time periods.
[0054] Furthermore, after obtaining the gold futures price forecast results, the following steps are also taken: comparing the gold futures price forecast results with the forecast results of the single Prophet model, as shown in Figure 2, and using RMSE, MAE and MAPE indicators to summarize the actual performance of different forecast models, as shown in Table 1.
[0055] Table 1
[0056] Model Auxiliary Predictors RMS EMA EMAPE Prophet None 5.4934 3.9526 3.4017% Prophet X1 1.2382 1.0178 0.8706% Prophet X1+X2 1.4932 1.1306 0.9758% LSTM None 3.8377 3.2430 2.7916% LSTM X2 7.7784 7.3924 6.3772% LSTM X1+X2 6.8145 6.2577 5.4058% Prophet-LSTM Prophet: None LSTM: None 5.3358 3.8824 3.3388% Prophet-LSTM Prophet: X1+X2 LSTM: None 1.4985 1.2207 1.0439% Prophet-LSTM Prophet: None LSTM: X1+X25.33913.88003.3373%Prophet-LSTMProphet: X1 LSTM: X21.17820.93720.8103%Prophet-LSTMProphet: X22.27932.14781.4816%Prophet-GRUProphet: X1 LSTM: X21.48621.17130.9898% surface
[0057] This invention discloses a futures price forecasting method combining time-lag analysis and a hybrid model. The Prophet model is a seasonal decomposition method based on time series, capable of handling time series data with strong seasonal patterns. By combining the Prophet model with the LSTM model, and analyzing and integrating the time lags of various economic indicators and financial variables, the data is input into the Prophet-LSTM model according to different time lags, achieving accurate capture of short-term fluctuations in gold futures prices. The Prophet-LSTM hybrid model not only utilizes the advantages of the Prophet model in processing time series data, but also fully leverages the capabilities of the LSTM model in capturing long-term dependencies and handling nonlinear patterns, while making the most reasonable allocation of different predictive factors. This invention fully considers the time-lag analysis of various price and macroeconomic environmental factor sequences that affect the trend of gold futures data. Compared with methods that do not consider predictive factors or other methods that incorporate predictive factors, the Prophet model, which considers immediate factors, and the LSTM model, which considers long-term factors, have the highest short-term prediction accuracy. Compared with traditional algorithms and other allocation methods, it can better capture data features on a monthly dimension, achieving more accurate short-term data prediction.
[0058] The above are merely preferred embodiments of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.
Claims
1. A futures price forecasting method combining time-delay analysis and a hybrid model, characterized in that, include: We acquire gold futures price datasets and time-series data of predictor factors, perform time lag analysis, and obtain predictor factors with optimal lag values less than the prediction length and those with optimal lag values greater than the prediction length. We input the predictor factors with optimal lag values less than the prediction length and their corresponding gold futures prices into the Prophet model within the hybrid model to obtain the fitted value, predicted value, and residual value of the gold price. We input the predictor factors with optimal lag values greater than the prediction length and the residual value of the gold price into the LSTM model within the hybrid model to obtain the residual predicted value of the gold price. Finally, we merge the predicted value from the Prophet model with the residual predicted value from the LSTM model to obtain the predicted result of the gold futures price.
2. The futures price prediction method combining time-delay analysis and hybrid models as described in claim 1, characterized in that, Obtaining predictors with optimal lag values less than the prediction length and predictors with optimal lag values greater than the prediction length involves: acquiring a gold futures price dataset and predictors; analyzing the correlation between gold futures prices and predictors to determine the degree of influence of different predictors on gold futures prices; performing optimal lag analysis on each predictor to obtain the optimal lag value; and analyzing the relationship between the optimal lag value and the prediction length of each predictor to obtain predictors with optimal lag values less than the prediction length and predictors with optimal lag values greater than the prediction length.
3. The futures price prediction method combining time-delay analysis and hybrid models as described in claim 1, characterized in that, Obtaining the fitted, predicted, and residual values of the gold price involves: inputting the predictor with the best lag value less than the prediction length and the corresponding gold futures price into the Prophet model in the hybrid model to predict the gold price; analyzing the predictor with the best lag value less than the prediction length and the corresponding gold futures price based on the Prophet model to generate the corresponding fitted and predicted values of the gold price; and calculating the residual value of the gold price by comparing the actual value of the gold price with the fitted value.
4. The futures price prediction method combining time-delay analysis and hybrid models as described in claim 2, characterized in that, Before obtaining the optimal lag value, it is necessary to analyze the correlation between the time series of gold futures prices and the time series of the predictor at different time lags to identify the potential relationship between the two series. The calculation is as follows: ,in, This represents the mean. and They are and The mean, and They are and standard deviation It is a time lag value.
5. The futures price prediction method combining time-delay analysis and hybrid models as described in claim 3, characterized in that, Modeling gold prices involves using a sliding window method, setting the window size to a preset size and the step size to a preset length. The Prophet model moves across the entire dataset, moving a preset number of data points each time. Each window contains a preset number of data points for training and prediction, and the performance of the Prophet model is observed over different time periods.
6. The futures price prediction method combining time-delay analysis and hybrid models as described in claim 1, characterized in that, The hybrid model is the Prophet-LSTM hybrid model.
7. The futures price prediction method combining time-delay analysis and hybrid models as described in claim 1, characterized in that, The predictive factors include other precious metal prices, crude oil futures prices, and macroeconomic indicators.
8. The futures price prediction method combining time-delay analysis and hybrid models as described in claim 3, characterized in that, The residual value of the gold price is the difference between the actual gold price and the fitted value obtained by training the Prophet model.