A short-term futures trading risk assessment system based on a mixture model

Abstract

The embodiment of the present application discloses a kind of short-term futures trading risk assessment systems based on mixed model, the embodiment of the present application realizes the intelligent identification and process control of transaction day by dynamic updating holiday library and multistage decision mechanism;Subsequently, multiple source data is normalized, interpolated and outlier processed, and a feature system covering cross-species correlation, multi-scale time lag, market microstructure and futures structure is constructed, and then key factors are screened by recursive feature elimination;On this basis, ARIMA model is used for benchmark prediction, and Prophet model is combined to correct residual, form a mixed prediction result, generate risk assessment data according to the prediction result;At the same time, introduce small perturbation mechanism and retry fault-tolerant strategy to ensure numerical calculation stability.The present application realizes the full-process automation from data preprocessing, feature engineering to model prediction, significantly improves the model robustness and practical efficiency.

Description

Technical Field

[0001] This invention relates to the field of artificial intelligence technology, specifically to a short-term futures trading risk assessment system based on a hybrid model. Background Technology

[0002] As a crucial component of the modern financial market system, the futures market's price fluctuations not only directly impact the trading decisions of hedgers and speculators but also profoundly influence price discovery and risk management across related industry chains. Therefore, timely short-term forecasting of futures prices has always been a core research topic in financial engineering and quantitative trading.

[0003] Currently, scholars and practitioners both domestically and internationally have conducted extensive research on futures price forecasting, resulting in various forecasting methods and technical systems. Traditional methods mainly include statistical models based on time series analysis, such as the Autoregressive Moving Average (ARIMA) model and the Generalized Autoregressive Conditional Heteroskedasticity (GARCH) model. These methods have certain explanatory power under small sample and linear assumptions, but they are difficult to effectively characterize nonlinear relationships, structural breakpoints, and external shocks in the market. With the development of machine learning technology, Support Vector Machines (SVM), Random Forests, Gradient Boosting Decision Trees (GBDT), and deep learning models (such as LSTM and GRU) have been gradually introduced into price forecasting. They excel in modeling feature nonlinearities, but they have high requirements for data quality, feature engineering, and model interpretability, and often ignore calendar effects and market microstructure information in trading time series.

[0004] In addition, existing forecasting systems still face several common challenges in practical applications: First, most models do not fully consider the impact of trading days and non-trading days (such as weekends and public holidays), leading to incorrect triggering of the forecasting process or data alignment chaos during non-trading periods; Second, feature construction relies heavily on technical indicators and historical prices, lacking systematic integration and dynamic selection of multi-dimensional information such as cross-product correlations, market sentiment, and futures-spot structure; Third, a single model often struggles to simultaneously capture linear trends, seasonal patterns, and responses to sudden events in time series, resulting in insufficient forecast robustness; Fourth, numerical computation stability issues can easily lead to model training failures in high-frequency or low-volatility market environments, affecting system availability.

[0005] To address the aforementioned issues, recent studies have attempted to improve prediction performance through model fusion, feature enhancement, or the introduction of external information. However, existing solutions often lack fully automated design across the entire process, from data preprocessing and feature construction to model optimization and result correction. Furthermore, systematic engineering solutions have not yet been developed for aspects such as holiday period identification, multi-scale feature lag, and low-variance data stability processing.

[0006] Therefore, there is an urgent need for a short-term futures price forecasting method that can integrate multi-source information, adapt to market calendar characteristics, have good numerical robustness, and can be automated, in order to improve system stability and practical applicability. Summary of the Invention

[0007] To address this, this invention provides a short-term futures trading risk assessment system based on a hybrid model, thereby solving the technical problems of poor model robustness and low automation in existing short-term futures price forecasting methods due to neglecting market calendar effects, relying on a single model, and insufficient feature construction.

[0008] To achieve the above objectives, the embodiments of the present invention provide the following technical solutions:

[0009] This invention provides a short-term futures trading risk assessment system based on a hybrid model, the system comprising:

[0010] Establish and dynamically update the holiday database. Use a date type determination algorithm to distinguish between trading days and non-trading days. If it is a trading day, continue with the subsequent steps; if it is a non-trading day, terminate the prediction process.

[0011] The raw data is cleaned and dailyized, and missing values ​​are filled by linear interpolation and backfilling. Outliers are processed using the Windsor method.

[0012] Based on the original data, cross-product correlation features, multi-scale time lag features, market microstructure features, and futures-spot structure features are constructed, and feature selection is performed through a recursive feature elimination algorithm.

[0013] After normalizing the selected features, the ARIMA-Prophet hybrid model is used to make short-term predictions of futures prices. The ARIMA model generates the baseline prediction, and the Prophet model corrects the residuals of the ARIMA prediction. Finally, the price prediction results for the next few trading days are output.

[0014] Furthermore, a holiday database is established and dynamically updated, and a date type determination algorithm is used to distinguish between trading days and non-trading days, including:

[0015] Establish an automated mechanism for dynamically updating the holiday database, detect and update holiday data in real time, and accurately identify date types through a three-level judgment logic to distinguish between weekends, holidays, and trading days.

[0016] Furthermore, outliers are processed using the Windsor method, which includes calculating upper and lower boundaries based on the 5th and 95th percentiles, and truncating or replacing values ​​that exceed the boundaries.

[0017] Furthermore, based on the original data, cross-product correlation features, multi-scale time lag features, market microstructure features, and futures-spot structure features are constructed, and feature selection is performed using a recursive feature elimination algorithm, including:

[0018] The time lag features include the short-term memory lag order set K={1,2,3,4,5}, the medium-term memory lag order set K={10,15}, and the long-term memory lag order set K={21}.

[0019] Furthermore, based on the original data, cross-product correlation features, multi-scale time lag features, market microstructure features, and futures-spot structure features are constructed, and feature selection is performed using a recursive feature elimination algorithm. This also includes:

[0020] The feature selection employs a recursive feature elimination algorithm, using a logistic regression model as the base evaluator, to recursively remove features with the smallest weights and retain the subset of features that have the greatest influence on the prediction target.

[0021] Furthermore, the ARIMA model uses the Akaike information criterion to automatically select the optimal (p,q) parameter combination, so that the model can be trained after passing the stationarity test.

[0022] Furthermore, after normalizing the selected features, a hybrid ARIMA-Prophet model is used to predict futures prices in the short term. The ARIMA model generates the baseline prediction, and the Prophet model corrects the residuals of the ARIMA prediction. The final output is the price prediction results for the next few trading days, which also includes:

[0023] Multi-factor parallel prediction of exogenous variables is performed to generate factor prediction values ​​for multiple future trading days, which are then used as input to the hybrid model.

[0024] Furthermore, the Prophet model uses the prediction residuals of the ARIMA model on the training set as the learning object, combines the prediction results of exogenous variables to predict the future residuals, and then weights and combines them with the ARIMA baseline prediction results to obtain the final prediction value.

[0025] Furthermore, the feature data is normalized before model training and denormalized after prediction to restore the results to their original dimensions.

[0026] According to a second aspect of the present invention, a short-term futures trading risk assessment system based on a hybrid model is provided, the system comprising:

[0027] The trading day detection and filtering module is used to establish and dynamically update the holiday database. It uses a date type determination algorithm to distinguish between trading days and non-trading days. If it is a trading day, it continues to execute subsequent steps; if it is a non-trading day, it terminates the prediction process.

[0028] The data preprocessing module is used to clean and diversify the raw data, perform linear interpolation and backfilling for missing values, and process outliers using the Windsor method.

[0029] Feature engineering is used to construct cross-product correlation features, multi-scale time lag features, market microstructure features, and futures-spot structure features based on raw data, and to select features through a recursive feature elimination algorithm.

[0030] The model training and prediction module is used to normalize the selected features and then use the ARIMA-Prophet hybrid model to make short-term predictions of futures prices. The ARIMA model generates the baseline prediction, and the Prophet model corrects the ARIMA prediction residuals. Finally, the module outputs the price prediction results for the next few trading days.

[0031] The embodiments of the present invention have the following advantages:

[0032] This invention achieves intelligent identification and process control of trading days through a dynamically updated holiday database and a multi-level judgment mechanism. Subsequently, multi-source data is normalized, imputed, and outlier-handled. A feature system covering cross-product correlations, multi-scale time lags, market microstructure, and futures-spot structure is constructed. Key factors are then screened through recursive feature elimination. Based on this, an ARIMA model is used for benchmark prediction, and the residuals are corrected using a Prophet model to form a hybrid prediction result. Risk assessment data is generated based on the prediction results. Simultaneously, a small perturbation mechanism and a retry fault-tolerance strategy are introduced to ensure the stability of numerical calculations. This invention achieves full automation from data preprocessing and feature engineering to model prediction, significantly improving model robustness and practical efficiency. Attached Figure Description

[0033] To more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are merely exemplary, and those skilled in the art can derive other embodiments based on the provided drawings without creative effort.

[0034] The structures, proportions, sizes, etc. illustrated in this specification are only for the purpose of assisting those skilled in the art in understanding and reading the content disclosed herein, and are not intended to limit the conditions under which the present invention can be implemented. Therefore, they have no substantial technical significance. Any modifications to the structure, changes in the proportions, or adjustments to the size, without affecting the effects and objectives that the present invention can produce, should still fall within the scope of the technical content disclosed in the present invention.

[0035] Figure 1 A schematic diagram of the logical structure of a short-term futures trading risk assessment system based on a hybrid model, provided for an embodiment of the present invention;

[0036] Figure 2 A flowchart illustrating a short-term futures trading risk assessment system based on a hybrid model, provided as an embodiment of the present invention;

[0037] Figure 3 is a schematic diagram of the first process of a short-term futures trading risk assessment system based on a hybrid model provided in an embodiment of the present invention;

[0038] Figure 4 This is a schematic diagram of the second process of a short-term futures trading risk assessment system based on a hybrid model, provided as an embodiment of the present invention. Detailed Implementation

[0039] The following specific embodiments illustrate the implementation of the present invention. Those skilled in the art can easily understand other advantages and effects of the present invention from the content disclosed in this specification. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0040] To address the technical problems of existing short-term futures price forecasting methods, which suffer from poor model robustness and low automation due to neglecting market calendar effects, relying on a single model, and insufficient feature construction.

[0041] refer to Figure 1 This invention discloses a short-term futures trading risk assessment system based on a hybrid model. The system includes: a trading day detection and screening module 1; a data preprocessing module 2; a feature engineering module 3; and a model training and prediction module 4.

[0042] refer to Figures 2 to 4The process involves establishing and dynamically updating a holiday database, using a date type determination algorithm to distinguish between trading days and non-trading days. If a date is a trading day, the subsequent steps continue; otherwise, the prediction process terminates. The raw data is cleaned, diversified to a daily scale, and missing values ​​are filled using linear interpolation and backfilling. Outliers are handled using Windsor normalization. Based on the raw data, cross-product correlation features, multi-scale time lag features, market microstructure features, and futures-spot structure features are constructed, and feature selection is performed using a recursive feature elimination algorithm. After normalizing the selected features, a hybrid ARIMA-Prophet model is used to predict futures prices in the short term. Risk assessment data is generated based on the prediction results, where the ARIMA model generates the baseline prediction, and the Prophet model corrects the ARIMA prediction residuals. Finally, the price prediction results for the next few trading days are output.

[0043] This invention establishes a comprehensive trading day detection and screening mechanism to ensure that the prediction process is executed only on valid trading days. The specific process is as follows:

[0044] 1. Holiday database dynamic update mechanism

[0045] Establish an automated holiday data maintenance system:

[0046] Version detection: Real-time detection of the latest version of the Chinese holiday database through command parsing.

[0047] Smart Update: Automatically installs upgrades when a new version is detected, ensuring the accuracy and timeliness of data during holidays.

[0048] Year Coverage Validation: After the update, verify the integrity of the holiday data for the current year to ensure that all transaction dates are covered.

[0049] Date type determination algorithm: It adopts a three-level determination logic to accurately identify the date type and divides the date into three categories: weekend, holiday and trading day.

[0050] Process control and execution management:

[0051] Implement differentiated process control based on date type determination results:

[0052] Non-trading day handling: When a weekend or holiday is detected, the system immediately terminates the subsequent prediction process and returns an error code.

[0053] Execution on Trading Day: After a trading day is confirmed as valid, the system continues to execute core processes such as data processing, feature engineering, and model prediction, and returns a success code.

[0054] Quarterly update strategy: Automatically trigger holiday database update checks between October and December each year to ensure the accuracy of the next year's transaction calendar.

[0055] 2. Data Preprocessing

[0056] Data normalization and daily processing: The raw data is cleaned and normalized. For weekly fundamental data (such as production, inventory, etc.), daily processing is performed using the forward imputation method to generate continuous daily time series, ensuring that all data remain consistent in time frequency.

[0057] Data integrity and continuity processing:

[0058] To ensure the continuity of the time series, a double imputation strategy is used for missing values ​​in the feature summary table:

[0059] Linear interpolation: Fills in missing values ​​in a time series using linear interpolation to ensure a smooth transition of data;

[0060] Backfill: For missing values ​​that still exist after linear interpolation, the backfill method is used to fill them with subsequent valid data to ensure the integrity of the dataset.

[0061] Outlier handling:

[0062] The Windsorization method is used to handle outliers in the data, based on and Set upper and lower boundaries to ensure the quality of training data. Among them: ( It is the 5th percentile. (95th percentile)

[0063]

[0064]

[0065] 3. Feature Engineering

[0066] 3.1 Feature Derivation and Construction

[0067] Cross-variety association characteristics:

[0068] Calculate the price differences and ratios between different futures contracts to capture the relative strength of the contracts and the transmission effect of the industrial chain.

[0069] Time delay characteristics:

[0070] By calculating the lagged values ​​of various prices, trading volumes, and open interest over multiple historical time windows, a multi-scale time-lag architecture is employed to capture:

[0071] Short-term memory (K={1,2,3,4,5}): Capturing recent market shocks and intraday effects.

[0072] Medium-term memory (K={10,15}): reflects cyclical fluctuations and trend continuation.

[0073] Long-term memory (K={21}): corresponding to monthly market structure and fundamental changes.

[0074] Definition of lag operator:

[0075] Let L be a lag operator that satisfies:

[0076]

[0077] Time delay feature set:

[0078] For each basic feature x_t, construct a time-delay feature set:

[0079]

[0080] Where K represents the set of subsequent orders, k represents the lag order, L is the lag operator, x_t is the observation of the time series at time t, and x_{tk} is the observation of the time series at time tk.

[0081] Market microstructure and capital sentiment characteristics:

[0082] Characteristics of volume-price relationship:

[0083] Volume-price correlation: measures the synergistic effect between increased trading volume and the direction of price changes.

[0084] Characteristics of fund flows:

[0085] Open interest change rate: reflects the net inflow and outflow of funds in futures funds.

[0086] Price behavior characteristics:

[0087] Intraday volatility: Indicates the degree of price fluctuation within a single day.

[0088] Closing / Closing Ratio: Reflects the strength of the closing price relative to the average price level of the day.

[0089] Liquidity characteristics:

[0090] Turnover rate: a measure of market trading activity and contract liquidity.

[0091] Futures-spot structure characteristics:

[0092] The futures market premium / discount structure and its changing trends are characterized by calculating the ratio (futures-spot ratio) and the difference (futures-spot difference) between futures prices and spot prices.

[0093] 3.2 Feature Selection

[0094] To avoid the curse of dimensionality and improve model efficiency, a recursive feature elimination algorithm is employed to filter all constructed features. This algorithm uses a logistic regression model as the base estimator and recursively removes features with the smallest weights, ultimately retaining the subset of features most influential on the prediction target (closing price). This achieves automated and intelligent feature selection, providing high-quality signal input for subsequent model predictions.

[0095] 4. Model Training and Prediction

[0096] 4.1 Data Transformation

[0097] Data normalization: Min-Max normalization is used to map features to the [0,1] interval, which improves the model's convergence stability.

[0098] For feature vectors Min-Max normalization is defined as:

[0099]

[0100] After the prediction is completed, the results need to be inversely normalized to the original dimensions:

[0101]

[0102] in, These are the normalized eigenvalues. The minimum value of the characteristic. For the maximum value of the characteristic, These are the original feature values.

[0103] 4.2 Model Parameter Training and Optimization

[0104] Model parameter training:

[0105] Stationarity test: The difference order d is determined by the ADF test to ensure the stationarity of the time series. The maximum difference order is limited to 2.

[0106] Parameter search space: The maximum search range for parameters p (autoregressive order) and q (moving average order) is set to 10, and the limit for specific high-frequency trading indicators is set to 5, balancing model complexity and overfitting risk.

[0107] Automated parameter selection mechanism:

[0108] The optimal combination of (p,q) parameters is automatically selected using the Akaike information criterion to minimize AIC and information loss.

[0109]

[0110]

[0111] in, This represents the maximum search range for the p-value. This represents the maximum search range for the value q.

[0112] Model performance evaluation:

[0113] The training set and the test set were divided in a 9:1 ratio.

[0114] MSE and RMASE are used as model evaluation metrics, where ε is a minimal constant to prevent division by zero.

[0115]

[0116]

[0117] in, For the true value, These are predicted values.

[0118] 4.3 Exogenous Variables and Indicator Prediction Module

[0119] Enhanced numerical stability:

[0120] Small perturbation mechanism: A uniformly distributed random perturbation is added to low-variance data to prevent singularity in the covariance matrix. For data vectors... The data after perturbation is as follows:

[0121] X_perturbed = X + Δ

[0122] The perturbation vector Δ satisfies:

[0123]

[0124] ε represents the disturbance amplitude, which is set to a default value. .

[0125] Dual detection mechanism: Detection is performed on both the original data and the normalized data; when the data variance... At this time, small perturbations are added to disrupt the complete consistency of the data, ensuring numerical stability.

[0126] Retry fault tolerance mechanism: Establish a retry mechanism of up to 3 times, and add small perturbations to automatically adjust and retrain when numerical calculation fails.

[0127] First retry:

[0128] Second retry:

[0129] Third retry:

[0130] Multi-factor parallel prediction:

[0131] Independent ARIMA modeling is performed on all important factors after feature selection to generate predicted values ​​for the next 5 trading days, providing exogenous variable inputs for subsequent hybrid models.

[0132] Metrics predictions for the ARIMA-Prophet hybrid architecture:

[0133] This invention employs a hybrid modeling framework combining ARIMA and Prophet to predict futures price indicators.

[0134] ARIMA baseline forecast:

[0135] An ARIMA model is constructed using the trained (p,d,q) parameters. The prediction results of exogenous variables are used as the model input to generate a benchmark price prediction for the next 5 trading days.

[0136] Prophet residual correction:

[0137] Extract the prediction residuals of the ARIMA model on the training set, construct the Prophet model to learn the temporal patterns of the residuals, and define the residuals as follows:

[0138]

[0139] The normalized exogenous variables are used as the regressor for Prophet to predict future residuals and correct the ARIMA results. The final prediction is a weighted combination of the ARIMA baseline prediction and the Prophet residual prediction.

[0140]

[0141] This hybrid model fully leverages ARIMA's strengths in linear time series modeling and Prophet's ability to capture complex seasonal and holiday effects, significantly improving the system's robustness.

[0142] It should be specifically noted that the core purpose of the method provided by this invention is to address the technical problem of existing futures forecasting relying on a single model and insufficient feature construction. The output forecast results are merely technical output signals generated by the model based on historical data and a set algorithm, and do not constitute any form of investment advice, buy / sell recommendation, or profit guarantee. The application of this method should be strictly limited to within the technical system, and users must combine it with other independent analytical methods and their own risk tolerance to make investment decisions. This invention assumes no responsibility for the accuracy, reliability, or actual investment results of the forecast results.

[0143] Although the present invention has been described in detail above with general descriptions and specific embodiments, modifications or improvements can be made to it, which will be obvious to those skilled in the art. Therefore, all such modifications or improvements made without departing from the spirit of the present invention fall within the scope of protection claimed by the present invention.

Claims

1. A short-term futures trading risk assessment system based on a hybrid model, characterized in that, The system includes: The trading day detection and filtering module is used to establish and dynamically update the holiday database. It uses a date type determination algorithm to distinguish between trading days and non-trading days. If it is a trading day, it continues to execute subsequent steps; if it is a non-trading day, it terminates the prediction process. The data preprocessing module is used to clean and diversify the raw data, perform linear interpolation and backfilling for missing values, and process outliers using the Windsor method. Feature engineering is used to construct cross-product correlation features, multi-scale time lag features, market microstructure features, and futures-spot structure features based on raw data, and to select features through a recursive feature elimination algorithm. The model training and prediction module is used to normalize the selected features and then use the ARIMA-Prophet hybrid model to make short-term predictions of futures prices. The ARIMA model generates the baseline prediction, and the Prophet model corrects the residual of the ARIMA prediction. Finally, it outputs the price prediction results for the next few trading days and generates risk assessment data based on the prediction results.

2. The short-term futures trading risk assessment system based on a hybrid model as described in claim 1, characterized in that, Establish and dynamically update a holiday database, and use a date type determination algorithm to distinguish between trading days and non-trading days, including: Establish an automated mechanism for dynamically updating the holiday database, detect and update holiday data in real time, and accurately identify date types through a three-level judgment logic to distinguish between weekends, holidays, and trading days.

3. The short-term futures trading risk assessment system based on a hybrid model as described in claim 1, characterized in that, Outliers are handled using the Windsor method, which includes calculating upper and lower boundaries based on the 5th and 95th percentiles, and truncating or replacing values ​​that exceed the boundaries.

4. The short-term futures trading risk assessment system based on a hybrid model as described in claim 1, characterized in that, Based on the original data, cross-product correlation features, multi-scale time lag features, market microstructure features, and futures-spot structure features are constructed. Feature selection is performed using a recursive feature elimination algorithm, including: The time lag features include the short-term memory lag order set K={1,2,3,4,5}, the medium-term memory lag order set K={10,15}, and the long-term memory lag order set K={21}.

5. The short-term futures trading risk assessment system based on a hybrid model as described in claim 4, characterized in that, Based on the original data, cross-product correlation features, multi-scale time lag features, market microstructure features, and futures-spot structure features are constructed. Feature selection is performed using a recursive feature elimination algorithm. The system also includes: The feature selection employs a recursive feature elimination algorithm, using a logistic regression model as the base evaluator, to recursively remove features with the smallest weights and retain the subset of features that have the greatest influence on the prediction target.

6. The short-term futures trading risk assessment system based on a hybrid model as described in claim 1, characterized in that, The ARIMA model uses the Akaike information criterion to automatically select the optimal (p,q) parameter combination, and the model is trained after passing the stationarity test.

7. The short-term futures trading risk assessment system based on a hybrid model as described in claim 6, characterized in that, After normalizing the selected features, an ARIMA-Prophet hybrid model is used to predict futures prices in the short term. The ARIMA model generates the baseline prediction, and the Prophet model corrects the residuals of the ARIMA prediction. The final output is the price prediction results for multiple trading days ahead, and also includes: Multi-factor parallel prediction of exogenous variables is performed to generate factor prediction values ​​for multiple future trading days, which are then used as input to the hybrid model.

8. The short-term futures trading risk assessment system based on a hybrid model as described in claim 7, characterized in that, The Prophet model uses the prediction residuals of the ARIMA model on the training set as the learning object, combines the prediction results of exogenous variables to predict the future residuals, and then weights and combines them with the ARIMA baseline prediction results to obtain the final prediction value.

9. The short-term futures trading risk assessment system based on a hybrid model as described in claim 8, characterized in that, Before model training, the feature data is normalized, and after prediction, it is denormalized to restore the result to its original dimensions.

Images