TWI824187B - Fund tracking system, fund tracking method and graphic user interface - Google Patents

Abstract

A fund tracking system, a fund tracking method and a graphic user interface are provided. The fund tracking method is used to track a target fund. The fund tracking method includes the following steps. Several ETF asset classes are obtained based a fund benchmark index of the target fund. Several representative ETFs are obtained based on the ETF asset classes. A simulated investment portfolio is generated based on the representative ETFs. Whether the simulated investment portfolio meets a verification condition is verified. If the simulated investment portfolio meets the verification condition, the simulated investment portfolio is outputted.

Description

Fund tracking system, fund tracking method and graphical user interface

This disclosure relates to a fund tracking system, a fund tracking method and a graphical user interface.

A mutual fund's returns depend on the return on the risky assets it holds. If we can predict the future trend of its risky assets, it will help predict the future returns of the fund. For example: OPEC's announcement of oil production cuts often helps oil-related indexes rise. Therefore, we can predict that funds that hold large amounts of oil assets will have a period of growth.

However, the fund holds risky assets across multiple industries and markets, and each asset holds a different proportion. It is difficult to predict the future trend of the fund based on the trend of a single asset.

In particular, funds disclose very little information to the outside world. They only publish comprehensive information such as companies with assets with larger weighting coefficients and their weighting coefficients, weighting coefficients classified by country, weighting coefficients classified by industry, etc. In addition, Fund Disclosure Week The period is longer, and the period for disclosing asset holding information is as short as one month, or even as long as a quarter or half a year.

Due to limited information, investors cannot effectively connect the information with the market to make trading decisions.

The present disclosure relates to a fund tracking system, a fund tracking system and a graphical user interface.

According to an embodiment of the present disclosure, a fund tracking method is proposed. The fund tracking method is used to track a target fund. The fund tracking method includes the following steps. Several index stock fund (Exchange Traded Fund, ETF) asset classes are obtained based on the fund benchmark index of one of the target funds. Get several representative ETFs based on these ETF asset classes. Based on these representative ETFs, a simulated investment portfolio is generated. Verify whether the simulated investment portfolio satisfies a verification condition. If the simulated investment portfolio meets the verification conditions, the simulated investment portfolio is output.

According to another embodiment of the present disclosure, a fund tracking system is provided. The fund tracking system is used to track a target fund. The fund tracking system includes a selection unit and a generation unit. The selection unit includes an asset class selector and an index stock fund (Exchange Traded Fund, ETF) selector. The asset class selector is used to obtain several ETF asset classes based on one of the target fund's fund benchmark indexes. The ETF selector is used to obtain several representative ETFs based on these ETF asset classes. The generation unit includes a combiner and a verifier. The combiner is used to generate a simulated portfolio based on these representative ETFs. Verify The device is used to verify whether the simulated investment portfolio meets a verification condition. If the simulated investment portfolio meets the verification conditions, the simulated investment portfolio is output.

According to yet another embodiment of the present disclosure, a patterned user interface is provided. The graphical user interface is used for a user to track a target fund. The graphical user interface includes a setting button and a fund prediction result button. The Set button is used to enter the target fund. Several index stock fund (Exchange Traded Fund, ETF) asset classes are obtained based on the fund benchmark index of one of the target funds. Several representative ETFs are obtained based on these ETF asset classes. A simulated portfolio is generated based on these representative ETFs. The Fund Forecast Result button is used to display the expected rise value of one of the simulated investment portfolios or a comprehensive trend.

In order to have a better understanding of the above and other aspects of the present disclosure, embodiments are given below and described in detail with reference to the accompanying drawings:

100:Fund tracking system

110:Selected unit

111:Asset Class Selector

112:ETF Selector

120:Generation unit

121:Combiner

122:Verifier

130: Trend prediction unit

900: Graphical user interface

Ci,C1,C2:ETF asset class

DCi:ETF Bullish Score

Dt: asset bullish score

DX: fund benchmark index

ECi, EC1, EC2: represents ETF

FC1: Short-term trend

FC2: Medium-term trend

FC3: Long-term trends

FC4: Comprehensive trends

FS, FS’: simulated portfolio

K1: Setting button

K2: Asset prediction result button

K3: ETF prediction result button

K4: Fund prediction result button

MD: deep learning model

P31: Fund Benchmark Index Retrieval Procedure

P32: Category retrieval procedure

P41: Weighting coefficient calculation program

P43: Weighting coefficient confirmation procedure

P44: Delete program

P45: Verification Procedure

P61: Market View Processor

P62: Feature extraction program

P63: Prediction program

P71: Natural Language Processing Program

P72: ETF expected value calculation program

P73: Fund growth expected value calculation program

P81: Sharpe value acquisition program

P82: Price retrieval procedure

P83: Feature regularization procedure

P84: Fill in 0 procedure

P85: Embedded feature extraction program

P86: Feature Integration Program

PEF: price embedded feature

PF: Price Features

Pij: selection probability

R0: risk-free interest rate

Rij: rating information ranking

S110, S120, S130, S140, S150, S160: steps

SCij: Number of times selected

SD: Financial Sentiment Dictionary

SF: Sharpe value feature

St: assets

T1, T2, T3, T4, T5, T6: time points

TB: Fund benchmark index and category correspondence table

TF: target fund

TX:Market View

UTV: rising expectations

WCi: weighting coefficient

Wij: selection probability weighting coefficient

Wit: configuration ratio

:空值

:null value

Figure 1 illustrates a block diagram of a fund tracking system according to one embodiment.

Figure 2 illustrates a flow chart of a fund tracking method according to an embodiment.

Figure 3 illustrates an example of the asset class selector executing step S110.

Figure 4 illustrates an example in which the combiner performs step S130 and the verifier performs step S140.

Figure 5 illustrates an example of executing the verification program P45.

Figure 6 illustrates an example of step S160 executed by the trend prediction unit.

Figure 7 illustrates an example of the market opinion processing program P61.

FIG. 8 illustrates an example of executing the feature extraction program P62.

Fig. 9 illustrates an example of executing the prediction program P63.

Figure 10 illustrates a schematic diagram of a patterned user interface according to an embodiment.

Index stock funds (Exchange Traded Fund, ETF) are composed of several risky assets. Compared with individual assets, ETFs conform to the composition characteristics of funds. In addition, compared with funds, ETFs have transparent asset allocation, fairly high trading information update frequency (for example, daily updates), and low tracking error ETFs can directly represent the index they track. Therefore, in this embodiment, researchers use ETFs to track funds to further predict the future direction of the fund.

Please refer to FIG. 1 , which illustrates a block diagram of a fund tracking system 100 according to one embodiment. The fund tracking system 100 is, for example, a server host, a computer, a cloud computing center or a smartphone. The fund tracking system 100 includes a selection unit 110, a generation unit 120 and a trend prediction unit 130. The selection unit 110, the generation unit 120 and the trend prediction unit 130 are, for example, a circuit, a chip, a circuit board, or a storage device that stores program code. The selection unit 110 includes an asset class selector 111 and an ETF selector 112 . The generation unit 120 includes a combiner 121 and a verifier 122 . The asset class selector 111, ETF selector 112, combiner 121 and validator 122 are, for example, a circuit, a chip, a circuit board, or a storage device storing program code. The fund tracking system 100 of this embodiment selects several representative ETF ECi (i=1,2,...) through the selection unit 110. The generation unit 120 then generates the simulated investment portfolio FS based on these representative ETFs ECi (i=1, 2,...). With the simulated portfolio FS, the target fund TF can be accurately tracked. Furthermore, the simulated investment portfolio FS can be used to predict the trend of the target fund TF. The following is a flow chart that explains the operation of each of the above components in detail.

Please refer to Figure 1 and Figure 2. Figure 2 illustrates a flow chart of a fund tracking method according to an embodiment. In step S110, the asset class selector 111 obtains several ETF asset classes Ci (i=1, 2,...) based on one of the fund benchmark indexes DX of the target fund TF. ETF asset categories Ci are, for example, industry categories, regional categories, fixed income categories, exchange rate categories, etc. Table 1 below also has examples illustrating the relationship between ETF asset categories Ci and the fund benchmark index DX.

Please refer to Figure 3, which illustrates an example of the asset class selector 111 executing step S110. After obtaining the target fund TF, the asset class selector 111 uses the fund benchmark index extraction program (benchmark index extraction) P31 to retrieve the fund benchmark index DX from the fund prospectus or the fund website. For example, the fund benchmark index DX of the stock fund "Fidelity Fund-Global Income Fund" is "MSCI ACWI NR USD"; the fund benchmark index DX of the bond fund "Fidelity Fund-USD Bond Fund" is "ICE BofA US LC Corp&Govt TR USD". The fund benchmark index DX corresponds to the investment objectives and areas of the target fund TF. If the fund benchmark index DX of two different target fund TFs is the same, the investment objectives and areas of the two target fund TFs will be the same.

Next, in the class extraction program (class extraction) P32, a lookup table (such as the fund benchmark index and class correspondence table TB) is used to obtain the ETF asset class Ci (i=1,2,...). Since the number of fund benchmark index DX is limited, it can be provided and maintained manually. For example, as shown in Table 1 below, it illustrates the fund benchmark index and category correspondence table TB. Different fund benchmark indexes DX may correspond to different numbers of ETF asset classes Ci (i=1,2,...). Since the changes in the fund benchmark index DX are limited, the fund benchmark index and category correspondence table TB can be easily maintained.

Then, in step S120 in Figure 2, the ETF selector 112 obtains several representative ETFs ECi (i=1,2,...) based on these ETF asset categories Ci (i=1,2,...) ). In this step, the representative ETF EC1 obtained is the best of several ETFs of the ETF asset class C1; the representative ETF EC2 obtained is the ETF asset class The best of several ETFs in C2; and so on. The selection of the best one is based on an ETF rating information ranking Rij to obtain each representative ETF ECi (i=1,2,...). The content of the rating information is, for example, "letter grade/fit value". The letter grade is a score for the Efficiency and Tradability section. The letter grade can objectively measure the performance of the ETF and the ease of buying and selling. The Fit value essentially measures the ETF's ability to occupy a larger market and can be selected based on investor needs. If you want to follow the market, you will choose the one with a higher Fit value; if you don't want to be the same as the broader market, you will choose the one with a lower Fit value. This embodiment takes selecting the one with a higher Fit value as an example.

For example, the rating information of a certain three ETFs is "A/85", "B/72", and "B/85". When comparing three ETF score information, they are first sorted by letter grade (A takes precedence over B, B takes precedence over C, and so on). Then sort by the Fit value (the one with the larger number takes precedence). Therefore, the order of the scoring information of these three ETFs is "A/85", "B/85", and "B/72", and their scoring information rankings Rij are 1, 2, and 3 respectively.

Please refer to Table 2 below, which illustrates the example of 11 representative ETF ECi (i=1,2,...) obtained based on the scoring information ranking Rij for 11 ETF asset classes Ci (i=1,2,...) ).

In one embodiment, in addition to directly selecting the representative ETF based on the best rating information ranking Rij, the selection can also be made with further reference to the number of times SCij is selected. For example, a selected representative ETF may be judged to be unsuitable and removed in the subsequent process of synthesizing the simulated portfolio FS. Therefore, when selecting representative ETFs in step S120, it is not suitable to always select those that are frequently removed.

The ETF selector 112 can select the representative ETF ECi (i=1,2,...) using the following equations (1) and (2).

In this article, i refers to different ETF asset classes and j refers to different ETFs within the same ETF asset class. The j-th ETF in the i-th ETF asset class Ci has a rating information ranking Rij, the number of selections SCij, a selection probability weighting coefficient Wij, and a selection probability Pij.

As shown in equation (1), the smaller the rating information ranking Rij, the greater the selection probability weighting coefficient Wij; the smaller the number of selections SCij, the greater the selection probability weighting coefficient Wij. Therefore, the better the rating information ranking Rij and the lower the number of selections SCij, the easier it is to be selected.

Next, in step S130 in Figure 2, the combiner 121 generates a simulated investment portfolio FS based on these representative ETFs ECi (i=1, 2,...). And in step S140 in Figure 2, the verifier 122 determines whether the simulated investment portfolio FS satisfies a verification condition.

The simulated investment portfolio FS is the product of these representative ETFs ECi (i=1,2,...) and the weighting coefficient WCi (i=1,2,...) (i.e. Σ _i WC _i * EC _i ). Please refer to FIG. 4 , which illustrates an example in which the combiner 121 executes step S130 and the verifier 122 executes step S140.

0)，代表沒有放空的情況。倘若組合器121在這些限制條件下無法計算出加權係數WCi(i=1,2,...)，則輸出空值

。 After the combiner 121 obtains the representative ETF ECi (i=1,2,...), it can calculate the weighting coefficient WCi (i=1,2,...) according to the regression model in the weighting coefficient calculation program P41. .). The regression model is, for example, a lasso regression model or a Ridge Regression model. In the process of calculating the weighting coefficient WCi (i=1,2,...), the combiner 121 must meet the following restrictions: the sum of all weighting coefficients WCi (i=1,2,...) is 1 (i.e. Σ _i WC _i =1). And any weighting coefficient WCi (i=1,2,...) is greater than or equal to 0 (that is, WC _i 1

0), indicating no short selling. If the combiner 121 cannot calculate the weighting coefficient WCi (i=1,2,...) under these constraints, it will output a null value

.

Then, in the weighting coefficient confirmation program P43, the combiner 121 confirms whether the weighting coefficients WCi (i=1, 2,...) are all greater than a predetermined weight value. When a certain plus When the weight coefficient is not greater than the predetermined weight value, it means that the corresponding representative ETF is not representative enough and needs to be deleted in the deletion procedure P44 and return to the weighting coefficient calculation procedure P41.

After successfully passing the weighting coefficient calculation procedure P41 and the weighting coefficient confirmation procedure P43, you can enter the verification procedure P45.

In the verification program P45, the verifier 122 verifies whether the simulated portfolio FS satisfies the verification conditions. The verification condition is that the similarity of the return change between the simulated investment portfolio FS and the target fund TF is less than a critical value. The verifier 122 uses, for example, the Kolmogorov-Smirnov test (K-S test) to analyze the approximation of the reward change.

Please refer to Figure 5, which illustrates an example of executing the verification program P45. As shown in the upper diagram of Figure 5, the combiner 121 uses the information from time points T1 to T2 to create a simulated investment portfolio FS at time point T2. Then, the verifier 122 can collect the return changes of the simulated investment portfolio FS and the return changes of the target fund TF at time points T2 to T3, and determine whether the similarity between the two is less than a critical value. After passing the verification, the simulated portfolio FS can be used to make predictions after time point T4.

In addition, as shown in the lower side of Figure 5, if you want to make predictions after time point T8, you can re-create a new simulated investment portfolio FS’ at time points T5~T6. If the simulated investment portfolio FS’ established at time point T6 can pass the verification at time points T6~T7. The simulated investment portfolio FS’ can be used to make predictions after time point T8.

Generally speaking, the investment percentage of target fund TF in the same industry category will not change much in a short period of time. In the two periods before and after (the shortest one month), the change in investment percentage in the same industry will usually not exceed 1%. Therefore, usually the simulated portfolio FS can maintain its effectiveness for a period of time.

，並重新挑選代表ETF(如第2圖步驟S140之結果為「否」時，回至步驟S120)。 As shown in Figure 4, if the verifier 122 verifies that the simulated investment portfolio FS meets the verification conditions, the simulated investment portfolio FS is output (for example, when the result of step S140 in Figure 2 is "Yes", step S150 is entered); if the verification The processor 122 verifies that the simulated investment portfolio FS does not meet the verification conditions, and outputs a null value.

, and re-select the representative ETF (for example, when the result of step S140 in Figure 2 is "No", return to step S120).

In step S150 of FIG. 2 , the verifier 122 outputs the simulated investment portfolio FS to the trend prediction unit 130 .

In step S160, the trend prediction unit 130 uses a deep learning model to predict the trend of the target fund TF based on the simulated investment portfolio FS. Please refer to FIG. 6 , which illustrates an example of step S160 performed by the trend prediction unit 130 . In the market view processing program P61, the expected rise value UTV of one of the simulated investment portfolios FS is calculated based on a market view.

Please refer to Figure 7, which illustrates an example of the market opinion processing program P61. In the natural language processing program P71, the market view TX is obtained, and the asset bullish score Dt(t=1, 2,...). t refers to different assets. For example, a certain market view TX is "Macquarie expects gold prices to approach US$2,000 per ounce." According to the financial sentiment dictionary SD, it can be analyzed that: "Macquarie" is the legal person, and "Gold Price" is the asset. St, "hopeful" is a 3-point opinion, and "approaching US$2,000 per ounce" is a 7-point opinion. The asset bullish score Dt is calculated to be 5.0 (i.e. (3+7)/2). For each asset St(t=1,2,...), the asset call score Dt(t=1,2,...) can be calculated in the same way.

Next, in the ETF expected value calculation program P72, the representative ETF ECi (i=1,2,...) is calculated based on the asset allocation ratio Wit of each representative ETF ECi (i=1,2,...) The ETF bullish score DCi (i=1,2,...). The ETF call score DCi is calculated based on the following formula (3), for example.

DC _i =Σ _t W _it * D _t ........................................ .............(3)

Then, in the fund's expected rise value calculation program P73, the expected rise value UTV of the simulated investment portfolio FS is calculated based on the ETF bullish score DCi (i=1,2,...). The expected increase value UTV is calculated based on the following formula (4), for example.

UTV=Σ _i WC _i * DC _i ..................................... ............(4)

Returning to Figure 6, in addition to obtaining the expected increase value UTV in the market view processing program P61, it is also necessary to obtain the price embedded feature PEF and Sharpe value feature SF in the feature extraction program P62. Please refer to Figure 8, which illustrates an example of executing the feature extraction program P62. In the Sharpe value extraction program P81, the Sharpe value characteristic SF is extracted based on the risk-free interest rate R0. In the price retrieval program P82, the price characteristics PF of the past period of time are retrieved. Furthermore, through the feature normalization program P83, the value of the price feature PF is moved to a range above 0. Then, through the 0-filling procedure P84, the price characteristic PF is filled with a numerical sequence of a certain preset length. Then, the embedded feature extraction program (embedded feature extraction) P85 is used to obtain the price embedded feature PEF. Finally, the Sharpe value feature SF and the price embedded feature PEF are integrated in the feature integration program P86. in training When building a model, all funds are trained together. Because the number of representative ETF ECi for each fund is different, some of the inputs in the training model are more (representing a greater number of ETF ECi) and some are less (representing a smaller number of ETF ECi). Deep learning cannot be performed if the number of inputs is inconsistent. All input quantities must be consistent in order to extract the same number of features. Therefore, in this embodiment, through the above-mentioned 0-filling procedure P84, the vacant characteristics of funds that represent a smaller number of ETF ECi are filled with 0 to perform the alignment operation.

Returning to Figure 6, after obtaining the expected increase value UTV, Sharpe value feature SF and price embedded feature PEF, the prediction program P63 can be used to predict the trend.

Please refer to Figure 9, which illustrates an example of executing the prediction program P63. In the prediction program P63, the deep learning model MD is mainly used to predict the trend of the target fund TF. The above-mentioned rising expected value UTV, Sharpe value feature SF and price embedding feature PEF can be input into the deep learning model MD to predict the short-term trend FC1, mid-term trend FC2 and long-term trend FC3 respectively. Then, the short-term trend FC1, mid-term trend FC2, long-term trend FC3, rising expectation value UTV, Sharpe value feature SF and price embedded feature PEF are input to the deep learning model MD to predict the comprehensive trend FC4. After having the comprehensive trend FC4, you can give suggestions on whether to buy or not.

In addition to the above implementation methods, when the expected increase value UTV is calculated, a recommendation on whether to buy can also be given directly based on the expected increase value UTV.

Through the above embodiment, the target fund TF can be accurately tracked, and further the simulated investment portfolio FS can be used to predict the trend of the target fund TF.

In addition, please refer to FIG. 10 , which illustrates a schematic diagram of a patterned user interface 900 according to an embodiment. For the convenience of users, directly use the smart It can be operated on a mobile phone, notebook computer, or desktop computer, and a graphical user interface 900 can be provided to execute the above fund tracking method. The graphical user interface 900 includes a setting button K1, an asset prediction result button K2, an ETF prediction result button K3, and a fund prediction result button K4. The setting button K1 is used to enter a window for the user to set the target fund TF. The asset prediction result button K2 is used to display the asset bullish scores Dt(t=1,2,...) of the above-mentioned various assets St(t=1,2,...). ETF prediction result button K3 is used to display the above-mentioned ETF bullish score DCi (i=1,2,...) representing ETF ECi (i=1,2,...). The fund prediction result button K4 is used to display the expected increase value UTV of the above-mentioned simulated investment portfolio FS, or the comprehensive trend FC4.

In summary, although the present disclosure has been disclosed in the above embodiments, they are not used to limit the present disclosure. Those with ordinary knowledge in the technical field to which this disclosure belongs can make various modifications and modifications without departing from the spirit and scope of this disclosure. Therefore, the protection scope of the present disclosure shall be subject to the scope of the appended patent application.

100:Fund tracking system

110:Selected unit

111:Asset Class Selector

112:ETF Selector

120:Generation unit

121:Combiner

122:Verifier

130: Trend prediction unit

Ci: ETF asset class

DX: fund benchmark index

ECi: represents ETF

FS: simulated portfolio

Pij: selection probability

Rij: rating information ranking

SCij: Number of times selected

Wij: selection probability weighting coefficient

Claims (16)

A fund tracking method is used to track a target fund. The fund tracking method is executed by a computer through a computer program product. The fund tracking method includes the following steps: acquiring a fund benchmark index based on the target fund, and The fund benchmark index corresponds to an investment goal and an area included in the target fund; based on the fund benchmark index, a plurality of index stock fund (Exchange Traded Fund, ETF) asset classes are obtained; based on the ETF asset classes, the corresponding assets for each A plurality of representative ETFs of the ETF asset class; a regression model is used to calculate a plurality of weighting coefficients related to the representative ETFs to generate a simulated investment portfolio following a first time interval, the simulated investment portfolio includes the representative ETFs and the Some weighting coefficients, wherein if one of the weighting coefficients is not greater than a predetermined weight value, and one of the weighting coefficients corresponds to the representative ETFs of the part, then delete the representative ETFs of the part; in the continuation In a second time interval after the first time interval, collect and determine whether the similarity between the return change of the simulated investment portfolio and the return change of the target fund is less than a critical value: if so, it is determined that the verification is passed And obtain a prediction result of the target fund, the prediction result follows a time point after the second time interval, if not, it is determined that the verification is not passed and the representative ETFs are re-selected; and In the step of obtaining the prediction result of the target fund, a deep learning model is used to learn to predict the trend of the target fund based on the market view and the financial sentiment dictionary to obtain the prediction result corresponding to the simulated investment portfolio. The fund tracking method as described in claim 1, wherein in the step of obtaining the ETF asset classes, the ETF asset classes are obtained based on a lookup table. The fund tracking method as described in claim 1, wherein in the step of obtaining the representative ETFs, each obtained representative ETF is the one with the best ranking information of a plurality of ETFs in each ETF asset class. The fund tracking method described in claim 1, wherein in the step of obtaining the representative ETFs, each representative ETF is obtained based on at least one ETF score information. The fund tracking method as described in claim 4, wherein in the step of obtaining the representative ETFs, each representative ETF is further obtained based on a number of selections, where the number of selections is the number of times it has been selected as a representative ETF. The fund tracking method as described in claim 1, wherein in the step of generating the simulated investment portfolio, a weighting coefficient of each representative ETF is obtained, the sum of the weighting coefficients is 1, and each weighting coefficient is greater than or equal to 0. The fund tracking method as described in claim 1, wherein the verification condition is that the similarity between the return change of the simulated investment portfolio and the target fund is less than a critical value. The fund tracking method as described in claim 1, wherein in the step of using the deep learning model to obtain the prediction result of the target fund, the prediction result is obtained based on at least a Sharpe value and a price embedded feature, wherein the price embedded feature The method is to retrieve a price feature in the past period of time, and then use an embedded feature extraction program to obtain the price embedded feature. A fund tracking system is used to track a target fund. The fund tracking system includes: a selection unit, including: an asset class selector for retrieving a fund benchmark index based on the target fund. The fund benchmark index is The target fund should include an investment objective and an area, and obtain a plurality of index stock fund (Exchange Traded Fund, ETF) asset classes corresponding to the fund's benchmark index; and an ETF selector for selecting based on the ETF asset classes. Obtain a plurality of representative ETFs; a generation unit, including: a combiner for calculating a plurality of weighting coefficients related to the representative ETFs according to a regression model to generate a simulated investment portfolio following a first time interval, the simulated investment The portfolio includes the representative ETFs and the weighting coefficients, where if one of the weighting coefficients is not greater than a predetermined weight value, and the other of the weighting coefficients The representative ETFs in a corresponding part of the corresponding part are deleted, and the representative ETFs in that part are deleted; and a verifier is used to collect and judge the simulated investment in a second time interval following the first time interval. Whether the degree of approximation between the change in the return of one of the portfolios and the change of the return of the target fund is less than a critical value: if so, it is determined that a prediction result of the target fund is obtained through verification, and the prediction result follows the second time interval At a later point in time, if not, it is determined that the representative ETFs have not passed the verification and the representative ETFs are re-selected; and a trend prediction unit is used to load a deep learning model for learning, and the deep learning model learns based on the market view and the financial sentiment. The dictionary predicts the trend of the target fund to obtain the prediction result corresponding to the simulated investment portfolio. The fund tracking system of claim 9, wherein the asset class selector obtains the ETF asset classes based on a lookup table. The fund tracking system of claim 9, wherein each of the representative ETFs obtained by the ETF selector is the one with the best ranking information for a plurality of ETFs in each of the ETF asset classes. The fund tracking system of claim 9, wherein the ETF selector obtains each representative ETF based on at least one ETF score information. The fund tracking system of claim 12, wherein the ETF selector further obtains each representative ETF based on a number of selections, where the number of selections is the number of times the representative ETF has been selected. The fund tracking system of claim 9, wherein the combiner obtains a weighting coefficient for each representative ETF, the sum of the weighting coefficients is 1, and each weighting coefficient is greater than or equal to 0. The fund tracking system as described in claim 9, wherein the verification condition is that the similarity between the return change of the simulated investment portfolio and the target fund is less than a critical value. The fund tracking system as described in claim 9, wherein the trend prediction unit obtains the prediction result based on at least a Sharpe value and a price embedded feature, wherein the price embedded feature is to capture a price feature in the past period of time, and then use An embedded feature extraction program obtains the price embedded feature.

Images