Intelligent decision-making system, device and method based on factor mining and machine learning

Abstract

The invention discloses an intelligent decision-making system, device and method based on factor mining and machine learning, and the method comprises the steps: data collection, factor mining, building of a training data set, building of a machine learning model, and output of decision-making suggestions. The invention relates to the two technical fields of factor mining and machine learning. Financial data are assisted to play an intelligent decision-making role on large revenue through factor mining, model construction and related operation. According to the method, on the basis of economics and machine learning theories, predictive factors with predictive ability and logic persuasion for the stock market return rate are mined and analyzed, and regularization processing is carried out on the predictive factors to reduce the weight of the factors with low importance, so that the stock market return rate is improved. The combination of predictive factors is realized and the weight is dynamically adjusted in an effective and steady manner, so that the stability of the result is improved, and finally, a large-disk return rate predictive value and shipping space suggestions are obtained.

Description

Technical Field

[0001] The present invention relates to the technical fields of factor mining and machine learning. Specifically, it relates to an intelligent decision-making system, device and method based on factor mining and machine learning. Background Art

[0002] For a long time, the ebb and flow of the stock market have shown great charm, fascinating the general public, professional investors and academic researchers. In the past few decades, more and more evidence has been accumulated, indicating that the overall returns of the stock market are predictable. Early studies focused on predictions based on past returns and found that short-term returns are positively autocorrelated with past returns, and long-term returns are negatively autocorrelated with past returns. The famous economist and expert in the field of financial economics, Eugene F. Fama, pointed out in his 1991 research results that the convincing power of the economic significance and statistical ability of these tests of return autocorrelation is insufficient. In subsequent studies of the stock market, indicators such as interest rates, dividend yields, and earnings price ratios were used to provide more powerful evidence for the predictability of returns.

[0003] Currently, with the continuous deepening of China's capital market, especially the secondary securities market, money market, and monetary policy reform, the marketization level of the stock market is getting higher and higher. With the implementation of the registration system, the increase in the cost of financial fraud, and the gradual unblocking of delisting channels, the stock market can increasingly reflect the economic fundamentals, and the linkage between stock returns and economic growth is getting stronger. The reform of the deposit and loan interest rate market and the introduction of the LPR have made China's interest rate transmission mechanism smoother, and the impact of changes in capital costs on the rise and fall of the stock market is becoming more obvious. The rigid repayment of bonds has gradually been broken, and the reference weight of credit ratings in the pricing of Chinese bonds is getting higher and higher, thus improving the predictive ability of credit spreads for market risks. All these make it feasible to predict the returns of A-shares.

[0004] Factor mining and factor combination are inseparable from the combination of mathematical analysis and economic logic. By using brute-force analysis of the correlation between market returns and factors with supercomputing power, undoubtedly a large number of effective factors will be obtained. As pointed out in the pioneering paper by Bates and Granger (1969), the prediction of factor combinations can outperform the prediction of individual factors themselves. However, this is often the result of overfitting. The performance within the sample is good, but the out-of-sample effect is often poor. Therefore, how to mine factors that are generally applicable to stock market prediction and give specific decision-making suggestions based on these factors is an important problem to be solved by the present invention. Summary of the Invention

[0005] The present invention provides an intelligent decision-making system, device and method based on factor mining and machine learning, which are used to solve at least one of the problems existing in the above-mentioned prior art.

[0006] The present invention provides an intelligent decision-making method based on factor mining and machine learning, which includes:

[0007] S01: Data collection

[0008] Collect financial data related to stock market returns through data access or data scraping and store it in a financial database. The financial data in the financial database is stored in chronological order according to the time nodes when it is generated;

[0009] S02: Factor mining

[0010] According to the influence strength, return prediction ability, the time sequence of data change and stock market return change, and the correlation degree with stock market returns, N prediction factors related to stock market returns are mined from the financial database, where N≥1 and is an integer;

[0011] S03: Construct a training data set

[0012] Extract data within a preset time range from the financial database and construct a training data set. The training data set includes the values of multiple prediction factors and the values of the rate of return in m time periods T1 to Tm within the prediction time range, where m≥1 and is an integer;

[0013] S04: Construct a machine learning model

[0014] Construct a machine learning model and train it with the data in the training data set;

[0015] S05: Output decision suggestions

[0016] Use the trained machine learning model to predict the overall market rate of return in a prediction time period and output decision suggestions. The decision suggestions include the predicted value of the overall market rate of return in the prediction time period and the position suggestions.

[0017] In a preferred embodiment, the multiple prediction factors include interest rate / exchange rate / currency factors, trade factors, price index factors, employment / wage factors, and fixed asset investment factors.

[0018] The interest rate / exchange rate / currency factors include: the 3-month RMB demand deposit rate, the 1-year RMB demand deposit rate, the 1-year treasury bond yield, the US dollar / yuan interest rate, the month-on-month growth rate of the US dollar / yuan interest rate, the real effective exchange rate index of the RMB, the money supply M2, the increment of social financing scale, and the newly added RMB loans.

[0019] The trade factors include: total retail sales of consumer goods, total retail sales of consumer goods of enterprises above designated size, and net export amount.

[0020] The price index factors include: consumer price index CPI, purchasing managers' index PMI, producer price index for industrial producers PPI, fixed-asset investment price index, and service production index.

[0021] The employment / wage factors include: newly increased urban employment and unemployment rate, average wage level of employed persons, and per capita disposable income of national residents.

[0022] The fixed-asset investment factors include: completed fixed-asset investment, land acquisition area for real estate, and construction scale of major products in the whole society this year.

[0023] In a preferred embodiment, constructing a machine learning model includes:

[0024] S11: Construct a regularization model, and the regularization model has a constant coefficient λ;

[0025] S12: Assign an initial value to the constant coefficient λ;

[0026] S13: Input the data in the training dataset into the regularization model, and calculate the predicted values of the market return rate for m time periods respectively i = 1, 2... m;

[0027] S14: Calculate the mean value of the actual market return rate for m time periods

[0028] S15: Calculate R-squared according to the following formula (1):

[0029]

[0030] where, y i represents the actual market return rate in the i-th time period;

[0031] S16: Adjust the value of the constant coefficient λ, and repeat S11~S15 to make R calculated according to formula (1) 2 maximum, and the value of the constant coefficient λ at this time is the optimal value;

[0032] S17: Output the regularization model corresponding to the optimal value.

[0033] In a preferred embodiment, the position suggestion for the prediction time period is obtained according to the following steps:

[0034] S21: Use the machine learning model to predict the overall market return rate for the prediction time period and the previous n time periods, where the lengths of the prediction time period and the n time periods are equal;

[0035] S22: Sort the n + 1 overall market return rates obtained from the prediction;

[0036] S23: Obtain the quantile corresponding to the overall market return rate of the prediction time period from the sorting result and compare it with a preset quantile value;

[0037] S24: When the quantile corresponding to the overall market return rate of the prediction time period is greater than the preset quantile value, the position recommendation is to increase the stock position; when the quantile corresponding to the overall market return rate of the prediction time period is less than the preset quantile value, the position recommendation is to reduce the stock position.

[0038] In a preferred embodiment, the lengths of the prediction time period and the n time periods are both one natural month or an integer number of natural months.

[0039] In a preferred embodiment, the preset quantile is calculated through the following steps:

[0040] S31: Select a time range and obtain the CSI 300 index for each month within the time range;

[0041] S32: Set j quantile estimations Q 1 ~Q j ;

[0042] S33: Calculate the cumulative returns of the CSI 300 index for each month within the time range corresponding to each quantile estimation respectively;

[0043] S34: Obtain the quantile estimation corresponding to the highest cumulative return as the preset quantile.

[0044] In a preferred embodiment, the intelligent decision-making method based on factor mining and machine learning further includes:

[0045] Visualization and interactive display: Display the prediction results of the overall market return rate on a human-computer interaction interface, which supports touch interaction, voice interaction, and gesture interaction. The prediction results of the overall market return rate include the short / medium / long-term prediction return rates of the CSI 300 and personalized investment recommendations.

[0046] In a preferred embodiment, the short / medium / long-term prediction return rates of the CSI 300 are the prediction return rates for the next quarter / year / three years respectively.

[0047] In a preferred embodiment, the basis for the personalized investment advice is at least one personalized dimension, and the personalized dimension includes the investor's age and investment risk preference.

[0048] In a preferred embodiment, the relationship between the personalized dimension and the corresponding position setting is as follows:

[0049]

[0050]

[0051] Calculate the initial recommended position according to the following formula (2):

[0052] Initial recommended position = M × First position setting × Second position setting (2)

[0053] M = 4.5 * Forecast value of the overall market return rate + 0.275 (3)

[0054] Wherein, M is the mapping value of the overall market return rate. Before calculating M, the forecast value of the overall market return rate in formula (3) is subjected to winsorization processing in the following manner:

[0055] When the forecast value of the overall market return rate is less than -0.05, let the forecast value of the overall market return rate be equal to -0.05.

[0056] When the forecast value of the overall market return rate is greater than 0.15, let the forecast value of the overall market return rate be equal to 0.15.

[0057] In a preferred embodiment, calculate the actual recommended position according to the following formula (4) and present it on the human-computer interaction interface:

[0058]

[0059] The present invention also provides an intelligent decision-making system based on factor mining and machine learning, which includes:

[0060] A data acquisition module, configured to collect financial data related to stock market returns by means of data access or data scraping and store it in a financial database, and the financial data in the financial database is stored in chronological order according to the time nodes when it is generated;

[0061] A factor mining module, configured to mine N prediction factors related to stock market returns from the financial database according to the influence strength, return prediction ability, time sequence of data change and stock market return change, and the correlation degree with stock market returns, where N ≥ 1 and is an integer;

[0062] A training dataset construction module, configured to extract data within a preset time range from the financial database and construct a training dataset, where the training dataset includes the values of multiple predictive factors and the values of the rate of return for m time periods T1 to Tm within the prediction time range, m≥1 and is an integer;

[0063] A machine learning model construction module, configured to construct a machine learning model and train it with the data in the training dataset;

[0064] A rate of return prediction module, configured to use the trained machine learning model to predict the overall market rate of return for a prediction time period and output decision-making suggestions, where the decision-making suggestions include the predicted value of the overall market rate of return for the prediction time period and the position suggestions.

[0065] The present invention also provides an intelligent decision-making device based on factor mining and machine learning, including a processor and a memory, where the memory stores instructions that can be executed by the processor, and when the instructions are executed by the processor, the above intelligent decision-making method is implemented.

[0066] Beneficial effects

[0067] The intelligent decision-making system, device, and method based on factor mining and machine learning provided by the present invention are supported by economic and machine learning theories, mine and analyze predictive factors that have the ability to predict the overall market rate of return of stocks and have logical persuasiveness, and perform regularization processing on the predictive factors to reduce the weights of factors with lower importance, adopt an effective and robust method to realize the combination of predictive factors and dynamically adjust the weights, improve the stability of the results, and finally obtain the predicted value of the overall market rate of return and the position suggestions. The present invention realizes the prediction of the future market rate of return to a certain extent, thereby guiding investors to increase the stock position when the market expected rate of return is high and reduce the stock position when the market expected rate of return is low, and then obtain greater profits and avoid losses. Therefore, the present invention has high commercial utilization value. In addition, the stock market is also a barometer of the real economy. When the rate of return improves or deteriorates, it must be because some macro factors improve or deteriorate. Using the present invention can also reflect the trend of the real economy in China and has great significance for warning market risks, especially in the context of the continuous compression of the overseas financing channels of Chinese enterprises. Description of the drawings

[0068] Figure 1 It is a flowchart of an intelligent decision-making method based on factor mining and machine learning according to an embodiment of the present invention;

[0069] Figure 2 It is a schematic diagram of a human-computer interaction interface according to an embodiment of the present invention;

[0070] Figure 3The architecture diagram of the intelligent decision-making system based on factor mining and machine learning according to an embodiment of the present invention. Detailed implementation manners

[0071] Embodiments of the present invention will be described in detail below. Examples of the embodiments are shown in the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all of the embodiments. These embodiments are merely exemplary and should not be construed as limiting the present invention. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts fall within the scope of protection of the present invention.

[0072] The development of the stock market in China has been going on for decades. The stock price fluctuates both in the short term and in the long term. The prediction of the stock market return has also been widely studied and concerned.

[0073] The difficulty in predicting the stock market return lies in that the stock price is affected by multiple factors, including not only national and international economic policies and financial policies, but also the development status and expectations of industries, etc. The three aspects of stocks, finance, and economy are closely related, interdependent, and mutually influential.

[0074] The financial market is a leading indicator of the economic situation. The stock market is not a zero-sum game. The profit of investors in the stock market comes not only from earning the losses of the counterparty, but more importantly, from the dividends of economic growth. When the economy is improving, the stock market tends to rise, and vice versa. Therefore, some indicators related to the economic fundamentals, especially leading indicators, will have the ability to predict the trend of the overall market; in addition to economic fundamental indicators, the stock market is also affected by many other macro factors, such as interest rates, CPI, etc. The change of interest rates reflects the change of investors' expectations for the future, and also affects the future cash flow discount rate and the corporate financing cost, so it has a great impact on the stock market; the overall stock market is similar to individual stocks. When the market valuation is too high, its high valuation is often unsustainable and leads to a decrease in future returns. When the overall market is undervalued, the possibility of obtaining higher returns in the future is greater; the market return and the volatility of the return in the past period also have the ability to predict future returns, because the market is an aggregation of individual stocks, and the prediction effects of these factors on individual stocks have been proven.

[0075] Figure 1 The flowchart of the intelligent decision-making method based on factor mining and machine learning according to an embodiment of the present invention. As Figure 1 shown, the intelligent decision-making method based on factor mining and machine learning provided by the present invention includes:

[0076] S01: Data collection

[0077] Collect financial data related to stock market returns by means of data access or data crawling and store them in a financial database, wherein the financial data in the financial database is stored in time series according to the time nodes when they were generated;

[0078] Data access can be, for example, access to a database interface provided by a third party. Data capture can be, for example, real-time capture of relevant data through a crawler. Data access and data capture are currently the main ways to obtain financial data. The present invention can also use other methods to obtain financial data, not limited to the above two methods. The acquired financial data is stored in a financial database according to the time node of generation, so as to facilitate subsequent steps.

[0079] S02: Factor Mining

[0080] Mining N prediction factors related to stock market returns from the financial database according to the influence, return prediction ability, the time sequence of data changes and stock market return changes, and the correlation with stock market returns, where N ≥ 1 and is an integer;

[0081] S03: Build training dataset

[0082] Extracting data within a preset time range from the financial database and constructing a training data set, the training data set including values ​​of multiple prediction factors and values ​​of yields in m time periods T1 to Tm within the prediction time range, where m≥1 and is an integer;

[0083] S04: Build a machine learning model

[0084] Construct a machine learning model and train it using the data in the training dataset;

[0085] S05: Output decision suggestions

[0086] The trained machine learning model is used to predict the market yield for a forecast time period and output decision recommendations, which include the forecast value of the market yield for the forecast time period and position recommendations.

[0087] Multiple predictors for the forecast period are fed into the trained machine learning model to obtain decision recommendations.

[0088] In one embodiment of the present invention, the plurality of prediction factors include interest rate / exchange rate / currency factor, trade factor, price index factor, employment / wage factor and fixed asset investment factor.

[0089] Interest rate / exchange rate / currency factors include: RMB 3-month demand deposit rate, RMB 1-year demand deposit rate, 1-year Treasury bond yield, USD / CNY exchange rate, month-on-month growth rate of USD / CNY exchange rate, real effective exchange rate index of RMB, money supply M2, increment of social financing scale, and newly added RMB loans,

[0090] Trade factors include: total retail sales of consumer goods, total retail sales of consumer goods of enterprises above designated size, and net export amount,

[0091] Price index factors include: consumer price index CPI, purchasing managers' index PMI, producer price index PPI for industrial producers, fixed asset investment price index, and service production index,

[0092] Employment / wage factors include: newly increased urban employment and unemployment rate, average wage level of employed persons, and per capita disposable income of national residents,

[0093] Fixed asset investment factors include: completed fixed asset investment, floor area of land purchased for real estate, and construction scale of major products in the whole society this year.

[0094] The above multiple prediction factors are factors with a relatively high degree of closeness to finance, stocks, and the economy. Over time, there may be other factors incorporated into the prediction factors, and some factors may no longer be effective prediction factors. The present invention is not limited to the limited several factors listed above.

[0095] In an embodiment of the present invention, S04 constructing a machine learning model includes:

[0096] S11: Construct a regularization model, and the regularization model has a constant coefficient λ;

[0097] The regularization model here is the regularized linear regression model, which adds a regularization term to the linear model to prevent overfitting of the model. Generally, there are L1 regularization and L2 regularization. The L1 regularization of linear regression is usually called Lasso regression. The difference between it and general linear regression is that a L1 regularization term is added to the loss function. The L1 regularization term has a constant coefficient λ to adjust the weight of the mean square error term and the regularization term of the loss function. The specific loss function expression of the Lasso regression regularization model is as follows:

[0098]

[0099] Where y is the predicted value, Xw is the true value, and w i is the coefficient of X;

[0100] S12: Assign an initial value to the constant coefficient λ;

[0101] The constant coefficient λ needs to be optimized to enhance the generalization ability of the model. Here, an initial value is preset for the constant coefficient λ first.

[0102] S13: Input the data in the training dataset into the regularization model, and calculate the predicted values of the market index return for m time periods respectively i = 1, 2... m;

[0103] Here, the m time periods are, for example, m natural months.

[0104] S14: Calculate the mean of the actual market index returns for m time periods

[0105] S15: Calculate R-squared according to the following formula (1):

[0106]

[0107] where, y i represents the actual market index return in the i-th time period;

[0108] S16: Adjust the value of the constant coefficient λ, and repeat S11 - S15 to make R 2 calculated according to formula (1) the largest. At this time, the value of the constant coefficient λ is the optimal value;

[0109] R 2 The constant coefficient λ when it is the largest is the final optimization result.

[0110] S17: Output the regularization model corresponding to this optimal value.

[0111] In formula (1) above, the numerator on the right side of the equal sign is all the errors predicted by the trained regularization model. The denominator below represents that if a blind guess is made, the result is the average of y. If the denominator is 0, it means that the trained regularization model is about the same as a blind guess. If the denominator is 1, it means that the regularization model fits without error. If the denominator is between 0 and 1, it represents the quality of the regularization model. If the denominator is negative, it means that the obtained regularization model is worse than a blind guess.

[0112] In an embodiment of the present invention, the position recommendation for the prediction time period is obtained according to the following steps:

[0113] S21: Use the machine learning model to predict the market index returns for the prediction time period and the previous n time periods, and the durations of the prediction time period and the n time periods are equal;

[0114] The duration of the prediction period and the n time periods is generally one natural month or an integer number of natural months, such as one month, one quarter (3 natural months), one year, three years, etc. Generally speaking, if it is necessary to predict the overall market return after one month, the duration taken here is one month; if it is necessary to predict the overall market return after one quarter, the duration taken here is one quarter, and so on.

[0115] S22: Sort the n + 1 overall market returns obtained by prediction;

[0116] S23: Obtain the quantile corresponding to the overall market return of the prediction period from the sorting result and compare it with a preset quantile value;

[0117] S24: When the quantile corresponding to the overall market return of the prediction period is greater than the preset quantile value, the position recommendation is to increase the stock position; when the quantile corresponding to the overall market return of the prediction period is less than the preset quantile value, the position recommendation is to reduce the stock position.

[0118] In a preferred embodiment, the preset quantile is calculated through the following steps:

[0119] S31: Select a time range and obtain the CSI 300 index for each month within the time range;

[0120] S32: Set j quantile estimations Q 1 ~Q j ;

[0121] Here, the larger the number of j, the finer the range of the quantile estimations Q 1 ~Q j and the higher the accuracy of the calculated preset quantile.

[0122] S33: Calculate the cumulative returns of the CSI 300 index for each month within the time range corresponding to each quantile estimation respectively;

[0123] In the case where the quantile estimation is a definite value, a prediction value of the overall market return can be calculated for each month. According to the strategy in step S24 above, decide whether to increase the stock position or reduce the stock position, and calculate the cumulative return for each month within the time range accordingly. For the convenience of calculation, increasing the stock position here can be calculated as full position, for example, and reducing the stock position can be calculated as liquidating the position, for example.

[0124] Calculate the cumulative returns corresponding to each quantile estimation in turn.

[0125] S34: Obtain the quantile estimation corresponding to the highest cumulative return as the preset quantile.

[0126] As Figure 2The figure shows a schematic diagram of a human - machine interaction interface according to an embodiment of the present invention. In an embodiment of the present invention, the intelligent decision - making method based on factor mining and machine learning further includes:

[0127] Visual interactive display: The predicted results of the market index return rate are displayed on a human - machine interaction interface. The human - machine interaction interface supports touch interaction, voice interaction, and gesture interaction. The predicted results of the market index return rate include the short / medium / long - term predicted return rates of the CSI 300 and personalized investment recommendations.

[0128] As Figure 2 shown, the human - machine interaction interface includes the predicted return rates of the CSI 300 for the next quarter / year / three years and the actual recommended positions (holding ratios) obtained according to two personalized dimensions: the age of the investor and the investment risk preference.

[0129] In an embodiment of the present invention, the relationship between the personalized dimension and the corresponding position setting is as follows:

[0130]

[0131]

[0132] The basis for the above - mentioned position setting is that generally, the lower the age of the investor, the stronger the risk - bearing ability, and aggressive investors tend to have higher positions.

[0133] Calculate the initial recommended position according to the following formula (2):

[0134] Initial recommended position = M×First position setting×Second position setting (2)

[0135] M = 4.5 * Predicted value of market index return rate+0.275 (3)

[0136] Among them, M is the mapped value of the market index return rate. Before calculating M, the predicted value of the market index return rate in formula (3) is trimmed as follows:

[0137] When the predicted value of the market index return rate is less than - 0.05, let the predicted value of the market index return rate be equal to - 0.05.

[0138] When the predicted value of the market index return rate is greater than 0.15, let the predicted value of the market index return rate be equal to 0.15.

[0139] Using the method of the present invention, in most cases, the predicted value of the market index return rate obtained is between [-0.05, 0.15]. This result is after trimming extreme values. After calculating according to the above formula (3), the range of the mapped value of the market index return rate is limited between [0.05, 0.95].

[0140] According to the above calculation method, when the predicted value of the overall market return rate is greater than or equal to the maximum value of 0.15, for an aggressive investor aged 30, his initial recommended position is 0.95×1×1. When the predicted value of the overall market return rate is less than or equal to the minimum value of -0.05, for a conservative investor aged 60, his initial recommended position is 0.05×1 / 4×1 / 3. For any investor, regardless of the overall market return rate, the initial recommended position is always between 0.05×1 / 4×1 / 3 and 0.95×1×1.

[0141] In an embodiment of the present invention, the actual recommended position is calculated according to the following formula (4) and presented on the man-machine interaction interface:

[0142]

[0143] The actual recommended position is the Figure 2 holding ratio in, which is a normalization process of the initial recommended position. After calculation according to the above formula (4), the actual recommended position is between 5% and 95%.

[0144] As Figure 3 shown is the architecture diagram of the intelligent decision-making system based on factor mining and machine learning in an embodiment of the present invention. As Figure 3 shown, the intelligent decision-making system based on factor mining and machine learning provided by the present invention includes:

[0145] The present invention also provides an intelligent decision-making system based on factor mining and machine learning, which includes:

[0146] A data acquisition module, configured to collect financial data related to stock market returns through data access or data scraping and store it in a financial database. The financial data in the financial database is stored in chronological order according to the time nodes of generation;

[0147] A factor mining module, configured to mine N prediction factors related to stock market returns from the financial database according to the influence strength, return prediction ability, time sequence of data change and stock market return change, and the correlation degree with stock market returns, where N≥1 and is an integer;

[0148] A training data set construction module, configured to extract data within a preset time range from the financial database and construct a training data set. The training data set includes the values of multiple prediction factors and the values of return rates in m time periods T1 to Tm within the prediction time range, where m≥1 and is an integer;

[0149] A machine learning model construction module, configured to construct a machine learning model and train it with the data in the training data set;

[0150] A return rate prediction module is used to predict the overall market return rate of a prediction time period by using a trained machine learning model and output decision suggestions. The decision suggestions include the predicted value of the overall market return rate of the prediction time period and the position suggestions.

[0151] The present invention also provides an intelligent decision-making device based on factor mining and machine learning, including a processor and a memory. The memory stores instructions that can be executed by the processor. When the instructions are executed by the processor, the above intelligent decision-making method is implemented.

[0152] The intelligent decision-making system, device and method based on factor mining and machine learning provided by the present invention are supported by economic and machine learning theories, mine and analyze prediction factors that have the ability to predict the overall market return rate of stocks and have logical persuasiveness, and perform regularization processing on the prediction factors to reduce the weights of factors with lower importance. An effective and robust method is adopted to combine the prediction factors and dynamically adjust the weights, improving the stability of the results. Finally, the predicted value of the overall market return rate and the position suggestions are obtained. The present invention realizes the prediction of the future market return rate to a certain extent, thereby guiding investors to increase the stock position when the market expected return rate is high and reduce the stock position when the market expected return rate is low, and then obtaining greater profits and avoiding losses. Thus, it can be seen that the present invention has high commercial utilization value. In addition, the stock market is also a barometer of the real economy. When the return rate improves or deteriorates, it must be because some macro factors improve or deteriorate. Using the present invention can also reflect the trend of China's real economy and has great significance for warning market risks, especially in the context of the continuous compression of the foreign financing channels of Chinese enterprises.

[0153] Those of ordinary skill in the art can understand that the drawings are only schematic diagrams of an embodiment, and the modules or processes in the drawings are not necessarily essential for implementing the present invention, and the "in one embodiment" in the text does not necessarily refer to the same embodiment but one of all possible embodiments.

[0154] Those of ordinary skill in the art can understand that the modules in the device in the embodiment can be distributed in the device of the embodiment according to the description of the embodiment, or can be correspondingly changed and located in one or more devices different from the present embodiment. The modules of the above embodiment can be combined into one module, or can be further split into multiple sub-modules.

Claims

1. An intelligent decision-making method based on factor mining and machine learning, characterized in that: include: S01: Data Collection Collect financial data related to stock market returns by means of data access or data crawling and store them in a financial database, wherein the financial data in the financial database is stored in time series according to the time nodes when they were generated; S02: Factor Mining Mining N prediction factors related to stock market returns from the financial database according to the influence, return prediction ability, the time sequence of data changes and stock market return changes, and the correlation with stock market returns, where N ≥ 1 and is an integer; S03: Build training dataset Extracting data within a preset time range from the financial database and constructing a training data set, the training data set including values ​​of multiple prediction factors and values ​​of yields in m time periods T1 to Tm within the prediction time range, where m≥1 and is an integer; S04: Build a machine learning model Construct a machine learning model and train it using the data in the training dataset; S05: Output decision suggestions The trained machine learning model is used to predict the market yield for a forecast time period and output decision recommendations, which include the forecast value of the market yield for the forecast time period and position recommendations.

2. The intelligent decision-making method based on factor mining and machine learning according to claim 1, characterized in that: Multiple predictive factors include interest rate / exchange rate / currency factors, trade factors, price index factors, employment / wage factors, and fixed asset investment factors. Interest rate / exchange rate / currency factors include: RMB 3-month demand deposit rate, RMB 1-year demand deposit rate, 1-year treasury bond yield, USD / RMB interest rate, USD / RMB interest rate month-on-month growth rate, RMB real effective exchange rate index, money supply M2, social financing scale increment and new RMB loan amount. Trade factors include: total retail sales of consumer goods, total retail sales of consumer goods of enterprises above designated size, and net export amount. Price index factors include: Consumer Price Index (CPI), Purchasing Managers Index (PMI), Producer Price Index (PPI), Fixed Asset Investment Price Index and Service Industry Production Index. Employment / wage factors include: the number of newly employed people in urban areas and the unemployment rate, the average wage level of employed people, and the per capita disposable income of residents nationwide. Fixed asset investment factors include: the amount of fixed asset investment completed, the area of ​​real estate land purchased, and the construction scale of major products in the whole society this year.

3. The intelligent decision-making method based on factor mining and machine learning according to claim 1, characterized in that: Building a machine learning model involves: S11: constructing a regularized model, wherein the regularized model has a constant coefficient λ; S12: assigning an initial value to the constant coefficient λ; S13: Input the data in the training data set into the regularized model, and calculate the predicted value of the market return rate for m time periods respectively S14: Calculate the mean of the actual market return over m time periods S15: Calculate R square according to the following formula (1): Among them, y i Represents the actual market return rate in the i-th time period; S16: Adjust the value of the constant coefficient λ, repeat S11 to S15, so that R calculated according to formula (1) 2 Maximum, at this time the value of the constant coefficient λ is the optimal value; S17: Output the regularized model corresponding to the optimal value.

4. The intelligent decision-making method based on factor mining and machine learning according to claim 1, characterized in that: The position recommendations for the forecast period are obtained by following the steps below: S21: using the machine learning model to predict the market yield of the forecast time period and the n time periods before it, where the forecast time period and the n time periods are equal in length; S22: Sort the predicted n+1 market returns; S23: Obtaining the quantile corresponding to the market return rate in the forecast time period from the sorting results and comparing it with a preset quantile value; S24: When the quantile corresponding to the market rate of return in the forecast period is greater than the preset quantile value, the position suggestion is to increase the stock position; when the quantile corresponding to the market rate of return in the forecast period is less than the preset quantile value, the position suggestion is to reduce the stock position; The duration of the predicted time period and the n time periods is one natural month or an integer number of natural months.

5. The intelligent decision-making method based on factor mining and machine learning according to claim 4 is characterized in that: The preset quantile is calculated by the following steps: S31: Select a time range and obtain the Shanghai and Shenzhen 300 Index for each month within the time range; S32: Set j quantile estimates Q1~Q j ; S33: Calculate the cumulative returns of the CSI 300 Index for each month within the time range corresponding to each quantile valuation respectively; S34: The quantile valuation corresponding to the highest cumulative return is obtained as the preset quantile.

6. The intelligent decision-making method based on factor mining and machine learning according to claim 1, characterized in that: Also includes: Visual interactive display: The forecast results of the market yield are displayed on a human-computer interaction interface, which supports touch interaction, voice interaction and gesture interaction. The forecast results of the market yield include the short-term / medium-term / long-term forecast yields of the CSI 300 and personalized investment advice; The short / medium / long term forecast yields of CSI 300 are the forecast yields for the next quarter / year / three years respectively.

7. The intelligent decision-making method based on factor mining and machine learning according to claim 6, characterized in that: The personalized investment advice is based on at least one personalized dimension, and the personalized dimension includes the investor's age and investment risk preference; The relationship between the personalized dimensions and the corresponding position settings is as follows: The initial recommended position is calculated according to the following formula (2): Initial recommended position = M × first position setting × second position setting (2) M = 4.5 * market return forecast + 0.275 (3) Where M is the mapping value of the market return rate. Before calculating M, the predicted value of the market return rate in formula (3) is shrinked as follows: When the predicted value of the market return is less than -0.05, set the predicted value of the market return to be equal to -0.

05. When the predicted value of the market return is greater than 0.15, set the predicted value of the market return equal to 0.

15.

8. The intelligent decision-making method based on factor mining and machine learning according to claim 7 is characterized in that: The actual recommended position is calculated according to the following formula (4) and presented on the human-computer interaction interface:

9. An intelligent decision-making system based on factor mining and machine learning, characterized in that: include: A data collection module, used to collect financial data related to stock market returns by means of data access or data capture and store the data in a financial database, wherein the financial data in the financial database is stored in time series according to the time nodes at which the data is generated; A factor mining module, for mining N prediction factors related to stock market returns from the financial database according to the influence, return prediction ability, the time sequence of data changes and stock market return changes, and the correlation with stock market returns, where N is ≥ 1 and is an integer; A training data set construction module, used to extract data within a preset time range from the financial database and construct a training data set, wherein the training data set includes values ​​of multiple prediction factors and values ​​of yields in m time periods T1 to Tm within the prediction time range, where m ≥ 1 and is an integer; A machine learning model building module, used to build a machine learning model and train it with the data in the training data set; The yield prediction module is used to use the trained machine learning model to predict the market yield for a prediction time period and output decision recommendations, which include the predicted value of the market yield for the prediction time period and position recommendations.

10. An intelligent decision-making device based on factor mining and machine learning, comprising a processor and a memory, characterized in that: The memory stores instructions that can be executed by the processor, and when the instructions are executed by the processor, the intelligent decision-making method described in any one of claims 1 to 9 is implemented.

Images