Investment portfolio construction method, device and equipment with advanced characteristic

Abstract

The invention discloses an investment portfolio construction method, device and equipment with an advanced characteristic, and relates to the technical field of stock market data processing.The method comprises the steps that factor index data are obtained, a deep learning network model is constructed, and the stock monthly frequency yield of the next period is predicted through the deep learning network model; the method comprises the following steps: constructing a multivariate LOG-HAR model based on high-frequency data, predicting a daily frequency, a weekly frequency and a monthly frequency realized covariance matrix of a future date by using the multivariate LOG-HAR model, introducing a predicted stock monthly frequency yield and a predicted monthly frequency realized covariance matrix into a Markovz model, and constructing an investment portfolio with an advanced characteristic; according to the method, the excess income is predicted through the predicted stock return rate and the monthly frequency realized covariance matrix, meanwhile, the high-frequency data is introduced to assist in predicting the monthly frequency realized covariance matrix, the return rate and the covariance matrix in the future time state are predicted on the whole, the uncertainty of the future market is greatly weakened, and the economic benefit of the market is improved. And the prediction accuracy is greatly improved.

Description

Technical Field

[0001] The present invention relates to the technical field of stock market data processing, and in particular to a method, device and equipment for constructing an investment portfolio with advanced characteristics. Background Art

[0002] As an investment portfolio strategy, asset allocation has undergone a significant evolution. Compared with traditional economics, modern portfolio theory is more in line with investor needs, supports diversified investment, and adapts to the development of the capital market. With the diversification of domestic investment tools such as stocks and futures, portfolio theory has become increasingly rich. The mean-variance model proposed by Markowitz (1952) provides the best allocation method. Although it is widely used, it has defects in parameter prediction. Berton (1980) further emphasized the importance of asset return estimation to the mean-variance framework, but accurate prediction is still challenging in practice, and slight errors may lead to significant impacts.

[0003] In the current portfolio theory, asset weight allocation is closely related to the risk relationship between assets (variance-covariance matrix) and expected rate of return. Accurate prediction of the covariance matrix and expected rate of return can optimize asset allocation and improve the Sharpe ratio and investment efficiency.

[0004] Traditional methods mainly rely on historical returns and covariance matrices, but investors are concerned about how to maximize investment returns and minimize risks in the future market environment. Therefore, when relying on historical returns and covariance matrices for predictions, they often ignore the uncertainty of the future market, resulting in lower prediction accuracy. Summary of the invention

[0005] The embodiments of the present invention provide a method, device and equipment for constructing an investment portfolio with advanced characteristics, which can solve the problem in the prior art that when relying on historical yields and covariance matrices for prediction, the uncertainty of the future market is often ignored, resulting in low prediction accuracy.

[0006] The embodiment of the present invention provides a method for constructing an investment portfolio with advanced characteristics, comprising the following steps:

[0007] Using daily stock market closing price data, obtain factor indicator data of closing price data;

[0008] Construct a deep learning network model, take the factor index data of the tth period as input, take the monthly yield value of the t+1th period as output, train the deep learning network model, and obtain the stock monthly yield prediction model; and obtain the predicted stock monthly yield based on the stock monthly yield prediction model;

[0009] According to the high-frequency data contained in the stock market closing price data within the preset time period, the daily realized covariance matrix is ​​obtained, and according to the daily realized covariance matrix, the weekly realized covariance matrix and the monthly realized covariance matrix are obtained;

[0010] A multivariate LOG-HAR model is constructed based on the high-frequency data contained in the stock market closing price data within a preset time period; the daily realized covariance matrix, weekly realized covariance matrix and monthly realized covariance matrix of the tth period are used as input, and the daily realized covariance matrix of the t+1th period is used as output to train the multivariate LOG-HAR model to obtain a realized covariance prediction model; and a predicted daily realized covariance matrix is ​​obtained based on the realized covariance prediction model, and a predicted monthly realized covariance matrix is ​​obtained based on the predicted daily realized covariance matrix;

[0011] The predicted monthly stock returns and the predicted monthly realized covariance matrix are input into the Markowitz model to construct an investment portfolio with forward-looking characteristics.

[0012] Preferably, the factor indicator data includes:

[0013] Stock monthly return factors, volatility index factors, market conditions index factors and financial index factors.

[0014] Preferably, the deep learning network model includes a first fully connected input layer, a second fully connected input layer, an LSTM recurrent network and a fully connected output layer;

[0015] The number of neurons in the first fully connected input layer is 6;

[0016] The number of neurons in the second fully connected input layer is 3;

[0017] The number of units in the LSTM recurrent network is 3;

[0018] The weight optimizer of the deep learning network model is the Adam optimizer;

[0019] The loss function of the deep learning network model is a mean square error loss function.

[0020] Preferably, obtaining the weekly realized covariance matrix and the monthly realized covariance matrix includes:

[0021] According to the high-frequency data, which is the closing price data of the Shanghai Stock Exchange 50 stocks within 5 minutes, the daily frequency realized covariance matrix is ​​obtained, and the acquisition equation is:

[0022]

[0023] Rj,t =lnp j,t -lnp j-1,t

[0024] Where: RCOV t represents the realized covariance matrix on the tth day; M represents the sample size on that trading day; R j,t represents the logarithmic return of the asset at time j on day t; p j,t represents the price at time j on day t;

[0025] Take the arithmetic average of the daily frequency realized covariance matrix within a week to obtain the weekly frequency realized covariance matrix;

[0026] The daily realized covariance matrix within a month is taken as the arithmetic average to obtain the monthly realized covariance matrix.

[0027] Preferably, the multivariate LOG-HAR model is:

[0028]

[0029] in: represents the error term.

[0030] Preferably, obtaining the predicted monthly frequency realized covariance matrix includes:

[0031] Input the daily frequency realized covariance matrix of period t+1 into the realized covariance prediction model to obtain the daily frequency realized covariance matrix of period t+2 after prediction, and cycle through them successively to obtain the predicted daily frequency realized covariance matrix within one month;

[0032] The predicted daily frequency realized covariance matrix within one month is calculated by arithmetic averaging to obtain the predicted monthly frequency realized covariance matrix.

[0033] The embodiment of the present invention further provides an investment portfolio construction device with advanced characteristics, comprising:

[0034] The data module is used to obtain the factor indicator data of the closing price data using the daily stock market closing price data;

[0035] The income module is used to construct a deep learning network model, which takes the factor indicator data of the tth period as input and the monthly yield value of the t+1th period as output, trains the deep learning network model, and obtains a stock monthly yield prediction model; and obtains the predicted stock monthly yield based on the stock monthly yield prediction model;

[0036] A matrix module is used to obtain a daily realized covariance matrix based on the high-frequency data contained in the stock market closing price data within a preset time period, and obtain a weekly realized covariance matrix and a monthly realized covariance matrix based on the daily realized covariance matrix;

[0037] A multivariate LOG-HAR model is constructed based on the high-frequency data contained in the stock market closing price data within a preset time period; the daily realized covariance matrix, weekly realized covariance matrix and monthly realized covariance matrix of the tth period are used as input, and the daily realized covariance matrix of the t+1th period is used as output to train the multivariate LOG-HAR model to obtain a realized covariance prediction model; and a predicted daily realized covariance matrix is ​​obtained based on the realized covariance prediction model, and a predicted monthly realized covariance matrix is ​​obtained based on the predicted daily realized covariance matrix;

[0038] The building module is used to input the predicted monthly stock returns and the predicted monthly realized covariance matrix into the Markowitz model to construct an investment portfolio with forward-looking characteristics.

[0039] An embodiment of the present invention further provides an electronic device, including a memory and a processor;

[0040] The memory is used to store computer programs;

[0041] The processor is used to implement the steps of the investment portfolio construction method with advanced characteristics as described above when executing the computer program stored in the memory.

[0042] The embodiment of the present invention provides a method, device and equipment for constructing an investment portfolio with advanced characteristics. Compared with the prior art, the beneficial effects thereof are as follows:

[0043] The present invention obtains factor index data through monthly stock data, constructs a deep learning network model according to the factor index data, and uses the deep learning network model to predict the monthly stock yield of the next period; constructs a multivariate LOG-HAR model based on high-frequency data, and uses the multivariate LOG-HAR model to predict the realized covariance matrix of the daily, weekly and monthly frequencies of future dates, introduces the predicted monthly stock yield of the next period and the predicted monthly realized covariance matrix of future dates into the Markowitz model, and constructs an investment portfolio with advanced characteristics; the present invention predicts excess returns through the predicted stock yield and the monthly realized covariance matrix, and introduces high-frequency data to assist in predicting the monthly realized covariance matrix. The overall method predicts the yield and covariance matrix under the future time state, greatly reduces the uncertainty of the future market, and greatly improves the accuracy of the prediction.

[0044] Furthermore, the present invention predicts the rate of return and covariance matrix at future moments, solves the lag problem of traditional investment portfolios, and improves the rate of return and stability. BRIEF DESCRIPTION OF THE DRAWINGS

[0045] Figure 1 A schematic diagram of the overall process of a method for constructing an investment portfolio with advanced characteristics provided by an embodiment of the present invention;

[0046] Figure 2 A schematic diagram of a flow chart of stock yield prediction of a method for constructing an investment portfolio with advanced characteristics provided by an embodiment of the present invention;

[0047] Figure 3 A schematic diagram of a realized covariance matrix prediction process of a method for constructing an investment portfolio with advanced characteristics provided by an embodiment of the present invention;

[0048] Figure 4 A schematic diagram of an investment portfolio strategy benefit with advanced characteristics of an investment portfolio construction method with advanced characteristics provided by an embodiment of the present invention;

[0049] Figure 5 A schematic diagram of monthly yield of an investment portfolio strategy with advanced characteristics of an investment portfolio construction method with advanced characteristics provided by an embodiment of the present invention;

[0050] Figure 6 A schematic diagram of Brinson analysis results of an investment portfolio strategy with advanced characteristics of an investment portfolio construction method with advanced characteristics provided in an embodiment of the present invention. DETAILED DESCRIPTION

[0051] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, the specific embodiments of the present invention are described in detail below in conjunction with the accompanying drawings. In the following description, many specific details are set forth to facilitate a full understanding of the present invention. However, the present invention can be implemented in many other ways different from those described herein, and those skilled in the art can make similar improvements without violating the connotation of the present invention, so the present invention is not limited by the specific embodiments disclosed below.

[0052] See also Figure 1 The embodiment of the present invention provides a method for constructing an investment portfolio with advanced characteristics, comprising the following steps:

[0053] Step 1: Based on the deep learning method, a stock monthly yield prediction model is constructed; for the stocks in the SSE 50, the factor data of the tth period is used as the input feature, and the monthly yield value of the t+1th period is used as the output target. After training, a stock monthly yield prediction model with advanced characteristics is obtained.

[0054] Step 2: Based on the high-frequency data, calculate the daily realized covariance matrix; based on the 5-minute closing price data of the stock, calculate the stock's return vector, and then calculate the daily frequency realized covariance matrix.

[0055] Step 3: Calculate the weekly and monthly realized covariance matrices; take the arithmetic average of the daily realized covariance matrices within a week to obtain the weekly realized covariance matrix; similarly, take the arithmetic average of the daily realized covariance matrices within a month to obtain the monthly realized covariance matrix.

[0056] Step 4: Construct a multivariate LOG-HAR model; take the daily realized covariance matrix, weekly realized covariance matrix and monthly realized covariance matrix of the tth period as input values, and the daily realized covariance matrix of the t+1th period as the output target, and after estimation, obtain a daily return frequency realized covariance prediction model with advanced characteristics.

[0057] Step 5: Predict the monthly frequency covariance matrix of period t+1; substitute the daily frequency realized covariance matrix obtained by the prediction of period t+1 into the prediction model obtained in step 4 to obtain the daily frequency realized covariance matrix of period t+2. Repeat this cycle to predict the daily frequency realized covariance matrices of t+2…t+19. Take the arithmetic mean of the 20 predicted daily frequency realized covariance matrices to obtain the predicted value of the monthly frequency covariance matrix.

[0058] Step 6: Based on the predicted monthly returns and the predicted monthly realized covariance matrix, construct Markowitz's "mean-variance" portfolio model, solve the model, and obtain the weights of the portfolio stocks with advanced characteristics.

[0059] In step one, using monthly Shanghai Composite 50 data, the factor data calculated include monthly stock yield factor, volatility index factor, market index factor and financial index factor; after obtaining the data set, the missing values ​​are eliminated and the StandardScaler() function is used for standardization.

[0060] In step 1, the deep learning model contains the first fully connected layer (6 neurons), the second fully connected layer (3 neurons), the third LSTM recurrent network (3 units), and a fully connected output layer, using the Adam optimizer and the mean squared error loss function.

[0061] In step 1, the model is set as:

[0062]

[0063] in: represents the predicted monthly rate of return for the t+1 period; Represents the historical monthly rate of return; Fi represents other factors, and the rest are correlation coefficients.

[0064] In step 2, the realized covariance matrix of the daily frequency is calculated as:

[0065]

[0066] R j,t =lnp j,t -lnp j-1,t

[0067] Where: RCOV t represents the realized covariance matrix on the tth day; M represents the sample size on that trading day; R j,t represents the logarithmic return of the asset at time j on day t; p j,t Represents the price at time j on day t.

[0068] In step 3, the weekly and monthly realized covariance matrices are calculated as follows:

[0069]

[0070] The actual number of trading days per month in my country's market is generally 20 days, and the number of trading days per week is generally 5 days.

[0071] In step 4, the multivariate LOG-HAR model is constructed as follows:

[0072]

[0073] Where: t+1,d represents the error term;

[0074] In step six, the Markowitz model of the advanced characteristics is constructed as follows:

[0075]

[0076] Where: X i represents the weight of the i-th stock; Σ represents the predicted monthly realized covariance; R represents the predicted monthly return.

[0077] The specific embodiments are:

[0078] Step 1: Specifically select the constituent stocks of the Shanghai Composite 50 Index as the stock pool, and select the monthly data from January 1, 2019 to December 31, 2022 as the data training sample. The data source is SuperMind; starting from the three major indicators of volatility indicators, market indicators and financial indicators, a total of 18 factors are constructed. The specific factor construction is shown in Table 1; the monthly frequency yield prediction process diagram is shown in Figure 2 shown.

[0079] Table 1 Factor structure

[0080]

[0081] Then, in the data preprocessing stage, ST and suspended stocks were eliminated, missing values ​​were eliminated, and standardized using the StandardScaler() function; the StandardScaler() function calculated the mean and standard deviation of the training set and standardized each factor data independently; then, the calculated mean and standard deviation were stored so that the same proportion could be used for standardization on the test set.

[0082] Build a deep learning model: the first fully connected layer (6 neurons), the second fully connected layer (3 neurons), the third LSTM recurrent network (3 units), and a fully connected output layer, using the Adam optimizer and the mean squared error loss function.

[0083] Among them, the first and second layers are used for dimensionality reduction, and the parameter values ​​are mainly referenced from the ideas of Fama's three-factor and six-factor models; the three-layer LSTM network uses a gating mechanism to finely control the flow and retention of information, so as to effectively capture the long-term dependencies in the time series; the last layer is the output layer, which outputs the predicted rate of return.

[0084] After the model is built, the model index parameters are determined and optimized, and the final confirmed model index parameters are shown in Table 2.

[0085] Table 2 Model index parameters

[0086] Learning Rate Time step Optimizer Loss Function Training cycle batch 0.01 3 Adam MSE 100 500

[0087] After training the model, the MSE on the training set dropped from 0.0921 to 0.0490, and the MSE on the test set was 0.0521, indicating that its fitting and prediction effects are good.

[0088] Step 4: Monthly frequency realized covariance matrix prediction process diagram as follows Figure 3 As shown; for the multivariate LOG-HAR model parameter estimation, when estimating, it is first necessary to use one day's high-frequency data to construct an n*n realized conditional covariance matrix Σ t , and decompose it into an orthogonal matrix B t and the diagonal matrix D t , for D t Logarithmization is performed to obtain the real symmetric matrix A t ; Its formula is:

[0089] A t =logm(Σ t )=Bt log(D t )B t '

[0090] Among them: B t The matrix is ​​an orthogonal matrix containing the realized covariance matrix Σ t The characteristic vector of t is a diagonal matrix containing the realized covariance matrix Σ t The eigenvalues ​​of the diagonal matrix D t Take the logarithm and get the real symmetric matrix A t ; Next, the real symmetric matrix A t Straighten, get The vector a t , substitute it into the LOG-HAR model for estimation, and the formula is:

[0091] a t =vech(A t )

[0092] After estimation, the inverse operation of matrix straightening is performed, which is Get n×n matrix Finally, the predicted realized covariance matrix Σ is obtained by exponentiation t .

[0093] Finally, in order to verify the effectiveness of the investment portfolio with advanced characteristics constructed by the strategy of the present invention, on the SuperMind platform, the stock is purchased in proportion to the stock weight of the advanced investment portfolio at the end of the month, and held until the end of the next month for readjustment; the backtest period is from January 1, 2023 to December 31, 2023; the initial capital is set to 1,000,000, the transaction fee is 0.02%, and the stock transaction slippage is 0.5%.

[0094] like Figure 4 As shown in FIG. 1 , it is a backtest profit chart of the investment portfolio strategy with advanced characteristics of the present invention; Figure 5 As shown, it is the backtest monthly return chart of the investment portfolio strategy with advanced characteristics of the present invention; it can be seen that the overall return of this strategy is higher than the benchmark, and when the benchmark return is in loss, the advanced investment portfolio strategy will not lose much; in historical data, the overall drawdown of the investment portfolio with advanced characteristics is small, and the overall volatility is not large, which shows that the investment portfolio with advanced characteristics has a lower risk level, the overall loss will not be too large, and it has better stability.

[0095] like Figure 6As shown, this is a Brinson analysis result diagram of the investment portfolio strategy with advanced characteristics of the present invention; it can be seen that the active return of this indicator is 7.15%, of which the position-related return occupies a large proportion; in the position-related returns, both the allocation return and the management return perform well, which shows that the ability of position management and internal stock management is relatively outstanding; in terms of management return, the return of individual stock selection is the most significant, indicating that the investment portfolio with advanced characteristics has excellent position management capabilities, pays more attention to stocks that can rise in the future, and fully reflects the core of the investment portfolio with advanced characteristics.

[0096] The present invention first calculates factor indicators through monthly stock data, and constructs a deep learning model based on these factors to predict the next monthly frequency yield; then, a multivariate LOG-HAR model is constructed using high-frequency data to predict the future daily, weekly and monthly frequencies and the realized covariance matrix; finally, the predicted monthly frequency yield and covariance matrix are substituted into the Markowitz model to construct an investment portfolio with advanced characteristics, and backtesting is performed; on the one hand, the present invention significantly improves the ability to obtain excess returns by adopting predicted data; on the other hand, the introduction of high-frequency data improves the accuracy of covariance matrix prediction, thereby optimizing the construction of the investment portfolio; compared with the traditional investment portfolio, the investment portfolio constructed by the present invention has more significant advantages in return rate and stability, solves the lag problem of the traditional investment portfolio, and better assists investment decision-making.

[0097] At the same time, on the one hand, the present invention improves the Markowitz "mean-variance" model by predicting stock returns and the covariance matrix between stocks, thereby more effectively obtaining excess returns; on the other hand, it overcomes the limitations of previous low-frequency data in covariance matrix prediction, and adopts the realized covariance matrix estimation method based on high-frequency data to more accurately construct an advanced characteristic investment portfolio.

[0098] The embodiment of the present invention further provides an investment portfolio construction device with advanced characteristics, comprising:

[0099] The data module is used to use the daily stock market closing price data to obtain the factor indicator data of the closing price data.

[0100] The income module is used to construct a deep learning network model. It takes the factor indicator data of the tth period as input and the monthly yield value of the t+1th period as output. The deep learning network model is trained to obtain a stock monthly yield prediction model; and the predicted stock monthly yield is obtained based on the stock monthly yield prediction model.

[0101] The matrix module is used to obtain the daily realized covariance matrix based on the high-frequency data contained in the stock market closing price data within a preset time period, and to obtain the weekly realized covariance matrix and the monthly realized covariance matrix based on the daily realized covariance matrix.

[0102] A multivariate LOG-HAR model is constructed based on the high-frequency data contained in the stock market closing price data within a preset time period; the daily realized covariance matrix, weekly realized covariance matrix and monthly realized covariance matrix of the tth period are used as input, and the daily realized covariance matrix of the t+1th period is used as output to train the multivariate LOG-HAR model to obtain a realized covariance prediction model; and the predicted daily realized covariance matrix is ​​obtained based on the realized covariance prediction model, and the predicted monthly realized covariance matrix is ​​obtained based on the predicted daily realized covariance matrix.

[0103] The building module is used to input the predicted monthly stock returns and the predicted monthly realized covariance matrix into the Markowitz model to construct an investment portfolio with forward-looking characteristics.

[0104] An embodiment of the present invention further provides an electronic device, including a memory and a processor.

[0105] The memory is used to store computer programs.

[0106] When the processor is used to execute the computer program stored in the memory, the steps of the above-mentioned investment portfolio construction method with advanced characteristics are implemented.

[0107] The above-mentioned embodiments only express several implementation methods of the present invention, and the descriptions thereof are relatively specific and detailed, but they cannot be understood as limiting the scope of the invention patent. It should be pointed out that, for ordinary technicians in this field, several variations and improvements can be made without departing from the concept of the present invention, and these all belong to the protection scope of the present invention. Therefore, the protection scope of the patent of the present invention shall be subject to the attached claims.

Claims

1. A method for constructing an investment portfolio with advanced characteristics, characterized in that: The following steps are involved: Using daily stock market closing price data, obtain factor indicator data of closing price data; Construct a deep learning network model, take the factor index data of the tth period as input, take the monthly yield value of the t+1th period as output, train the deep learning network model, and obtain the stock monthly yield prediction model; and obtain the predicted stock monthly yield based on the stock monthly yield prediction model; According to the high-frequency data contained in the stock market closing price data within the preset time period, the daily realized covariance matrix is ​​obtained, and according to the daily realized covariance matrix, the weekly realized covariance matrix and the monthly realized covariance matrix are obtained; A multivariate LOG-HAR model is constructed based on the high-frequency data contained in the stock market closing price data within a preset time period; the daily realized covariance matrix, weekly realized covariance matrix and monthly realized covariance matrix of the tth period are used as input, and the daily realized covariance matrix of the t+1th period is used as output to train the multivariate LOG-HAR model to obtain a realized covariance prediction model; and a predicted daily realized covariance matrix is ​​obtained based on the realized covariance prediction model, and a predicted monthly realized covariance matrix is ​​obtained based on the predicted daily realized covariance matrix; The predicted monthly stock returns and the predicted monthly realized covariance matrix are input into the Markowitz model to construct an investment portfolio with forward-looking characteristics.

2. A method for constructing an investment portfolio with advanced characteristics according to claim 1, characterized in that: The factor indicator data include: Stock monthly return factors, volatility index factors, market conditions index factors and financial index factors.

3. A method for constructing an investment portfolio with advanced characteristics according to claim 1, characterized in that: The deep learning network model includes a first fully connected input layer, a second fully connected input layer, an LSTM recurrent network and a fully connected output layer; The number of neurons in the first fully connected input layer is 6; The number of neurons in the second fully connected input layer is 3; The number of units in the LSTM recurrent network is 3; The weight optimizer of the deep learning network model is an Adam optimizer; The loss function of the deep learning network model is a mean square error loss function.

4. A method for constructing an investment portfolio with advanced characteristics according to claim 1, characterized in that: The step of obtaining the weekly realized covariance matrix and the monthly realized covariance matrix includes: According to the high-frequency data, which is the closing price data of the Shanghai Stock Exchange 50 stocks within 5 minutes, the daily frequency realized covariance matrix is ​​obtained, and the acquisition equation is: R j,t =lnp j,t -lnp j-1,t Where: RCOV t represents the realized covariance matrix on the tth day; M represents the sample size on that trading day; R j,t represents the logarithmic return of the asset at time j on day t; p j,t represents the price at time j on day t; Take the arithmetic average of the daily frequency realized covariance matrix within a week to obtain the weekly frequency realized covariance matrix; The daily realized covariance matrix within a month is taken as the arithmetic average to obtain the monthly realized covariance matrix.

5. A method for constructing an investment portfolio with advanced characteristics according to claim 4, characterized in that: The multivariate LOG-HAR model is: in: represents the error term.

6. A method for constructing an investment portfolio with advanced characteristics according to claim 1, characterized in that: The method for obtaining the predicted monthly frequency realized covariance matrix includes: Input the daily frequency realized covariance matrix of period t+1 into the realized covariance prediction model to obtain the daily frequency realized covariance matrix of period t+2 after prediction, and cycle through them successively to obtain the predicted daily frequency realized covariance matrix within one month; The predicted daily frequency realized covariance matrix within one month is calculated by taking the arithmetic mean to obtain the predicted monthly frequency realized covariance matrix.

7. A portfolio construction device with advanced characteristics, characterized in that: include: The data module is used to obtain the factor indicator data of the closing price data using the daily stock market closing price data; The income module is used to construct a deep learning network model, which takes the factor indicator data of the tth period as input and the monthly yield value of the t+1th period as output, trains the deep learning network model, and obtains a stock monthly yield prediction model; and obtains the predicted stock monthly yield based on the stock monthly yield prediction model; A matrix module is used to obtain a daily realized covariance matrix based on the high-frequency data contained in the stock market closing price data within a preset time period, and obtain a weekly realized covariance matrix and a monthly realized covariance matrix based on the daily realized covariance matrix; A multivariate LOG-HAR model is constructed based on the high-frequency data contained in the stock market closing price data within a preset time period; the daily realized covariance matrix, weekly realized covariance matrix and monthly realized covariance matrix of the tth period are used as input, and the daily realized covariance matrix of the t+1th period is used as output to train the multivariate LOG-HAR model to obtain a realized covariance prediction model; and a predicted daily realized covariance matrix is ​​obtained based on the realized covariance prediction model, and a predicted monthly realized covariance matrix is ​​obtained based on the predicted daily realized covariance matrix; The building module is used to input the predicted monthly stock returns and the predicted monthly realized covariance matrix into the Markowitz model to construct an investment portfolio with forward-looking characteristics.

8. An electronic device, comprising: Memory and processor; The memory is used to store computer programs; The processor is used to implement the steps of the method for constructing an investment portfolio with advanced characteristics as claimed in any one of claims 1 to 6 when executing the computer program stored in the memory.

Images