WO2022004957A1

WO2022004957A1 - Investment advisory method and apparatus therefor

Info

Publication number: WO2022004957A1
Application number: PCT/KR2020/015026
Authority: WO
Inventors: 김한샘
Original assignee: 유한책임회사 알케미랩
Priority date: 2020-07-02
Filing date: 2020-10-30
Publication date: 2022-01-06

Abstract

Disclosed are an investment advisory method and an apparatus therefor. The investment advisory apparatus: calculates a plurality of predicted prices by training a predictive model that outputs a predicted price by using time series data of financial information, and then repeatedly inputting, a predefined number of times, time series data at a current point in time into the predictive model that has been completely trained; identifies the distribution of the plurality of predicted prices; and calculates an investment weight on the basis of the ratio between a first area of the higher distribution and a second area of the lower distribution that have been identified in the distribution on the basis of a current price.

Description

Investment advisory method and device

An embodiment of the present invention relates to an investment advisory method and apparatus for financial assets such as stocks, bonds, and bitcoin, and more particularly, to a method and apparatus for advising investment proportion of financial assets.

In general, investment advice on stocks, etc., is provided by experts with know-how and information on high-yielding investment stocks. Recently, there are robo-advisors that plan investment strategies through advanced algorithms and big data. However, since the general robo-advisor algorithm has a large difference in performance and investment results depending on the change in parameter values, various errors including overfitting-bias can be made. Therefore, there is a need for a more general type of automated investment algorithm that is not affected by parameters.

The technical problem to be achieved by the present invention is to provide a method and an apparatus for advising the best investment weight within the range of future uncertainty inferred by an artificial intelligence model.

In order to achieve the above technical problem, an example of an investment advisory method according to an embodiment of the present invention includes: learning a predictive model for outputting a forecast price using time series data of financial information; calculating a plurality of predictive values by repeatedly inputting the time series data at the current time into the learned predictive model a predefined number of times; identifying a distribution of the plurality of predictors; and calculating the investment weight based on the ratio of the first area of the upper distribution and the second area of the lower distribution determined based on the current price in the distribution.

In order to achieve the above technical problem, an example of an investment advisory apparatus according to an embodiment of the present invention includes: a learning unit for learning a predictive model for outputting a predictive price using time series data of financial information; a predictor for calculating a plurality of predictive values by repeatedly inputting the time series data at the current time into the predictive model a predefined number of times; a distribution grasping unit which grasps a distribution of the plurality of predictors; and an investment weighting calculator configured to calculate the investment weight based on the ratio of the first area of the upper distribution to the second area of the lower distribution, which is determined based on the current price in the distribution.

According to an embodiment of the present invention, it is possible to calculate the investment weight of various financial assets, such as stocks, bonds, bitcoins, futures, and options, and provide them to investors. In another embodiment, a more accurate investment weight can be obtained by using not only time series data of financial information but also augmented data obtained by processing time series data. In another embodiment, it is possible to advise investors on the proportion of investment tailored to the investor by reflecting the investor's return on the balance.

1 is a view showing an example of a schematic structure of an investment advisory system according to an embodiment of the present invention;

2 is a diagram illustrating an example of time series data used for learning a predictive model according to an embodiment of the present invention;

3 is a diagram illustrating an example of a method for learning a predictive model using time series data according to an embodiment of the present invention;

4 is a view showing an example of a predictor obtained using a predictive model according to an embodiment of the present invention;

5 is a diagram illustrating an example of a method of augmenting learning data in order to increase the accuracy of a predictive model according to an embodiment of the present invention;

6 is a diagram illustrating an example of a distribution of predictors obtained through a predictive model according to an embodiment of the present invention;

7 is a view showing an example of a method for calculating the investment weight according to an embodiment of the present invention;

8 is a view showing an example of a method of correcting the investment weight to the investor balance return according to an embodiment of the present invention;

9 is a flowchart illustrating an example of an investment advisory method according to an embodiment of the present invention, and;

10 is a diagram showing the configuration of an example of an investment advisory apparatus according to an embodiment of the present invention.

Hereinafter, an investment advisory method and apparatus according to an embodiment of the present invention will be described in detail with reference to the accompanying drawings.

1 is a diagram illustrating an example of a schematic structure of an investment advisory system according to an embodiment of the present invention.

Referring to FIG. 1 , the investment advisory apparatus 100 may be connected to at least one financial system 110 and at least one user terminal 120 .

The financial system 110 is a system that provides information on various financial assets, such as stocks, bonds, futures, and bitcoins. The investment advisory apparatus 100 may collect various financial information through the financial system 110 to generate learning data necessary for learning the predictive model. In another embodiment, the investment advisory apparatus 100 may receive learning data directly from the manager. In this case, the financial system 110 may be omitted.

The user terminal 120 may access the investment advisory device 100 to receive advice on investment weight and the like. For example, when the user requests the investment advisory device 100 for advice on the investment weight of a specific stock item through his/her user terminal 120 , the investment advisory device 100 uses the prediction model for the requested stock It can be provided to the user terminal 120 by calculating the investment weight for the.

The investment advisory apparatus 100 may be implemented as a general computer, a server, or a cloud system. In another embodiment, the investment advisory apparatus 100 may be implemented as an application and mounted on the user terminal 120 to be performed.

2 is a diagram illustrating an example of time series data used for learning a predictive model according to an embodiment of the present invention.

Referring to FIG. 2 , the time series data 200 for financial information includes at least one item 210 , and each item includes values a ₁₁ to a _mn for a plurality of time points 220 . The time series data 200 may include various information items that can directly or indirectly affect the price of a financial asset (eg, stock, bond, bitcoin, etc.) for which the investment weight is to be calculated. For example, time series data for stock A includes various items such as stock index, starting price and closing price, selling/buying status by subject, exchange rate, interest rate, and government bond interest rate, and each item may have values for a certain period of time.

The time series data 200 may be directly generated by an administrator and input to the investment advisory apparatus 100 . For example, an administrator can create time series data in Excel format. In another embodiment, the investment advisory apparatus 100 may access the financial system 110 and scrape necessary information to automatically generate the time series data 200 . To this end, in the investment advisory apparatus 100, address information (eg, URL address) of the financial system in which financial information is stored and the type of information to be scraped from the financial system (eg, start and close prices for each stock item) are predefined. may have been

The time series data 200 for financial information may be used as learning data of the predictive model as it is, or outliers may be removed from the time series data 200 through various conventional pre-processing processes and used as learning data. Hereinafter, for convenience of explanation, it is assumed that the financial information time series data 200 of FIG. 2 is used as the learning data.

3 is a diagram illustrating an example of a method for learning a predictive model using time series data according to an embodiment of the present invention.

Referring to FIG. 3 , the prediction model 300 is an artificial intelligence model that receives financial information time series data (x) and outputs the price (ie, predictive price) (y) of a future financial asset based on this. For example, if FIG. 2 is time series data for stock A, the prediction model 300 may be an artificial intelligence model that outputs a forecast price for stock A.

The prediction model 300 may be implemented with various conventional artificial intelligence models, such as convolutional neural networks (CNNs). Also, the predictive model 300 may be trained by an unsupervised learning method. The AI model and the learning method of the AI model itself are already widely known technologies. This embodiment can apply various known AI models and learning methods, and is not limited to a specific AI model or a specific learning method.

The investment advisory apparatus 100 may learn a predictive model by repeating the process of inputting time series data of a certain section from the financial information time series data 200 as shown in FIG. 2 . For example, referring to FIG. 2 , the investment advisory apparatus 100 provides _{first input data composed of time series data of t 1} to t ₁₀ _{, second} input data composed of time series data of t 2 to t ₁₁ , t Input data composed of time series data in units of 10 intervals, such as third input data composed of time series data of ₃ to t _{12 intervals, may be input to the prediction model 300 .} A section unit constituting the input data may be set in various ways according to an embodiment. _{The predictive model 300 obtains the first predictive value at time t 11} for the first input data, compares the first predictor with _{the actual price at time t 11 in the} time series data, and adjusts the weights of the artificial intelligence model. process can be performed. By repeating this process for each input data, the predictive model 300 can be learned.

4 is a diagram illustrating an example of a predictor obtained by using a predictive model according to an embodiment of the present invention.

Referring to FIG. 4 , the investment advisory apparatus 100 may input time series data of financial information at the present time into the prediction model 300 to determine the predicted price of financial assets at a future time. Here, the prediction model 300 refers to a model that has been trained using the training data including the time series data of FIG. 2 .

For example, as in the example in FIG. 3 , when the prediction model 300 is an artificial intelligence model that receives time series data in units of 10 sections and outputs a forecast price, the investment advisory device 100 is By inputting the time series data in units of each section into the learned prediction model 300 , it is possible to determine the predictive value. Hereinafter, the time series data of the present time is defined as time series data from the present time point to a certain period in the past. Also, 'current time' means the most recent time of time series data input to the prediction model. Therefore, when time series data up to June 1, 2020 is input to the prediction model, the current time is June 1, 2020, and the forecast model outputs the forecast price after June 1, 2020.

The investment advisory apparatus 100 may obtain a plurality of forecast prices by repeatedly inputting the time series data of the current time into the prediction model 300 . For example, the investment advisory apparatus 100 may _{repeatedly input time series data of the current time point t n} into the prediction model 300 j times to obtain j predicted values for the time _{t n+1 time 420 .} Since the predictive model 300 does not always output the same predictive value even when time series data of the same current time is repeatedly input, some or all of the j predicted values may be different from each other.

In another embodiment, the investment advisory apparatus 100 _{inputs the time series data of the current time point (t n} ) to the prediction model 300 _{and not only the next time point (t n+1} ) but also a certain future section (t _n+1 ~ t _{n) +k} ) predictors (y _1(n+1) to y _j(n+k) ) can be obtained. For example, when time series data at the present time is input to the prediction model 300 , the prediction model 300 is not only the predictor value at time _{t n+1} _{(y 1(n+1)} ) but also t _n+1 ,.. It is possible to output the predictor value (y _1(n+2) ~y _1(n+k) _{) at the time .t n+k.} When the time series data at the present time is repeatedly input to the prediction model 300 j times _{, and a plurality of predictors for a certain section (eg, t n+1} ~ t _n+k ) are calculated through the prediction model 300 , , j*k predictors can be obtained as shown in FIG. 4 . In this case, the predictive model 300 may use the previously obtained predictor as input data. In other words, when the forecasters of t _{n + 1} time 420 calculated, investors sentence value t _{n + 1} to input a past time-series data of a period of time including forecasters at the time the predictive model 300 of t _{n + 2} time predictors can be obtained.

5 is a diagram illustrating an example of a method of augmenting learning data in order to increase the accuracy of a predictive model according to an embodiment of the present invention.

Referring to FIG. 5 , the investment advisory apparatus 100 generates a net graph 500 composed of at least one moving average line based on the time series data 200 as shown in FIG. 2 . The moving

averages

510 , 520 , and 530 may be moving averages for various sections, such as a 5-day moving average, a 20-day moving average, or a 60-day moving average. Although this embodiment shows examples of three different moving

average lines

510, 520, and 530 for convenience of explanation, the number of moving average lines may be set in various ways depending on the embodiment. As an example, the net graph 500 is one It may consist only of a moving average line.

The investment advisory apparatus 100 does not need to obtain a moving average line for all items of the time series data 200 , and items for which a moving average line is obtained from the time series data 200 may be predefined. For example, when the time series data 200 is data related to a specific stock, the investment advisory apparatus 100 may generate a moving average line of various sections with respect to the closing price of the specific stock in the time series data 200 .

The investment advisory apparatus 100 determines the distance (550,552,554) between the value (a) of each time point (t ₁ ~ t _n ) of the time series data 200 and each moving average line (510,520,530) and each time point (t ₁ ~t _n ) Find the slope (560,562,564) of the moving average at . This embodiment shows an example of the distance (550,552,554) between the value 540 of the time series data at time _{t a} and each moving average line (510, 520, 530) and the slope (560, 562, 564) of each moving average line at the time _{ta a.}

The investment advisory apparatus 100 generates augmented data including the distance and the slope of the moving average obtained at each time point. In addition, the investment advisory apparatus 100 uses the time series data and the augmented data of FIG. 2 as learning data of the prediction model 300 of FIG. 3 together.

6 is a diagram illustrating an example of a distribution of predictors obtained through a predictive model according to an embodiment of the present invention.

Referring to FIG. 6 , the predictor distribution 600 represents the distribution of the cumulative number of prediction values between the _{minimum predictor value b min} and the maximum predictor value b _{max .} 6 shows the shape of the normal distribution for convenience of explanation, but the shape of the distribution of the actual predictor may vary.

The investment advisory device 100 identifies the location of the current price 630 in the predicted price distribution 600, and calculates the ratio of the first area 610 of the upper distribution of the current price 630 and the second area 620 of the lower distribution. can figure out For example, if the prediction model 300 is a model for predicting the price of stock A, the current price 630 is the price of stock A at the present time.

The investment advisory apparatus 100 may determine a positive investment weight or a negative investment weight based on the ratio of the first area 610 to the second area 620 . For example, if the first area 610 is larger than the second area 620 (that is, (distribution average value b _avg − present price)>0), the investment advisory apparatus 100 sets the first area 610 and A positive investment ratio is calculated in proportion to the ratio of the second area 620 , and on the contrary, if the first area 610 is smaller than the second area 620 (that is, (distribution average value (b _avg ) - present price) <0), a negative investment weight proportional to the ratio of the first area 610 and the second area 620 may be calculated. Here, a positive investment weighting means increasing the investment weight of financial assets (eg, long position), and a negative investment weighting means reducing the investment weight of financial assets (eg, a short position). do. For accurate calculation of the investment weight proportional to the ratio of the first area 610 and the second area 620 , the investment weight graph shown in FIG. 7 may be used.

7 is a diagram illustrating an example of a method for calculating an investment weight according to an embodiment of the present invention.

6 and 7 together, the investment advisory apparatus 100 may define a relationship between the ratio of the first area 610 and the second area 620 and the investment weight 730 . For example, the relationship between the ratio of the first area 610 and the second area 620 and the investment weight 730 may be defined as a linear proportional relationship or may be defined as an S-curve relationship. can be defined.

In one embodiment, it is calculated using the Kelly formula (f = (bp-q)/b, where f is the investment weight, p is the win probability, q is the loss probability, and b is the net dividend rate), which is often used in investment. A relationship between the ratio of the first area 610 and the second area 620 and the investment weight 730 may be defined. For example, the investment advisory apparatus 100 applies the respective ratios of the first area and the second area to the total area in the distribution 600 of FIG. 6 to the win probability and the loss probability of the Kelly formula, respectively, and "investment ratio" (730) versus (distribution average value - present price) (740)" can be obtained. However, if a graph is drawn using only the Kelly formula, it becomes a graph that continues to increase without a lower limit or an upper limit, so a graph 700 as shown in FIG. 7 can be obtained by multiplying the sigmoid function so that the investment weight exists within a certain range. have.

In the example of FIG. 6 , since the difference between the current price 630 and the distribution average value (b _avg ) is ℓ, the investment advisory apparatus 100 is located in the graph 700 of FIG. The investment weight f% (720) can be obtained. If the prediction model 300 is a prediction model for stock A, the investment advisory apparatus 100 may recommend that the investment in stock A be increased by f% 720 from the total investment ratio. As another example, if the present price 630 in FIG. 6 _{exists in a place 750 that is separated by ℓ above the distribution average value b avg} , (distribution average value-present price) becomes negative and the investment weight -f in the graph of FIG. % (760) is calculated. In this case, the investment advisory apparatus 100 may recommend reducing the investment weight for the stock A from the total investment weight by f%.

The investment weight of FIG. 7 is an example of recommending the investment weight without considering the current situation of the investor. For example, if the return on balance of investor A and the return on balance of investor B are different, the advice on investment weight needs to be different accordingly. An example of this is shown in FIG. 8 .

8 is a diagram illustrating an example of a method of correcting an investment weight by an investor's balance return according to an embodiment of the present invention.

Referring to FIG. 8 , the relationship between the balance return 810 and the additional investment weight 820 is defined as a graph 800 . This embodiment considers the additional investment weight 840 only when the investor's balance return 830 is a negative number, and also sets the additional investment weight to be fixed when the balance return 840 is less than a certain level. The graph 800 of this embodiment is only an example for helping understanding, and the relationship between the balance return 810 and the additional investment weight 820 may be defined in various ways according to embodiments.

The investment advisory apparatus 100 calculates the additional investment weight 840 corresponding to the balance return 830 of the investor using the graph 800 of FIG. 8 and then corrects the investment weight 730 obtained in FIG. 7 based on this. do. For example, the investment advisory apparatus 100 may calculate and provide the value of “investment ratio * (1 + additional investment ratio)” as the final investment ratio for the investor.

9 is a flowchart illustrating an example of an investment advisory method according to an embodiment of the present invention.

Referring to FIG. 9 , the investment advisory apparatus 100 learns a predictive model using time series data on financial information ( S900 ). An example of time series data used as training data of a predictive model is shown in FIG. 2 . Also, an example of a learning method of a predictive model using time series data is shown in FIG. 3 . In another embodiment, the investment advisory apparatus 100 may train the predictive model by using the training data including the time series data of FIG. 2 and the salpin augmented data of FIG. 5 . For example, in FIG. 3 , the prediction model receives time series data and augmented data together.

When the learning of the predictive model is completed, the investment advisory apparatus 100 inputs the time series data of the current time into the predictive model to determine the predictive value of the financial asset (S910). At this time, the investment advisory apparatus 100 calculates a plurality of forecast prices by inputting the time series data of the current time into the prediction model a plurality of times. The investment advisory apparatus 100 may determine the forecast price at various points in the future through the prediction model as shown in FIG. 4 .

The investment advisory apparatus 100 identifies the distribution of a plurality of predictors (S920). For example, the investment advisory apparatus 100 may determine the cumulative distribution of the forecast price as shown in FIG. 6 . The investment advisory apparatus 100 calculates the investment weight by using the ratio of the first area of the upper distribution and the second area of the lower distribution, which is determined based on the current price in the distribution of the predicted price (S930). An example of calculating the investment weight is shown in FIG. 7 . As another example, the investment advisory apparatus 100 may calculate an investment weight suitable for the corresponding investor by reflecting the return on the investor's balance, an example of which is shown in FIG. 8 .

Referring to FIG. 10 , the investment advisory apparatus 100 includes a learning data generating unit 1000 , a learning unit 1010 , a prediction unit 1020 , a distribution grasping unit 1030 , an investment weight calculation unit 1040 , and a correction unit. (1050). According to an embodiment, the learning data generating unit 1000 and the correcting unit 1050 may be omitted.

The learning data generating unit 1000 generates time series data including at least one item of financial information as learning data. As an embodiment, the learning data generation unit 1000 may generate only the time series data of FIG. 2 as learning data or may generate the learning data by combining the augmented data and the time series data of FIG. 5 .

The learning unit 1010 learns a predictive model for outputting predictive prices for financial assets by using time series data of financial information. For example, if the learning data consists of augmented data and time series data, the learning unit 1010 may learn the predictive model by inputting the augmented data and time series data together into the predictive model.

The predictor 1020 calculates a plurality of predictor values by repeatedly inputting the time series data of the current time into the predictive model a predefined number of times. An example of a method of calculating a predictor is shown in FIG. 4 .

The distribution determining unit 1030 grasps the distribution of the plurality of predictors obtained by the predicting unit 1020 . An example of the distribution of predictors is shown in FIG. 6 .

The investment weight calculation unit 1040 calculates the investment weight based on the ratio of the first area of the upper distribution and the second area of the lower distribution determined based on the current price in the distribution of the predicted price obtained by the distribution grasping unit 1030 . An example of a method of calculating the investment weight is shown in FIG. 7 .

The correction unit 1050 corrects the investment weight by reflecting the additional investment weight corresponding to the return on the balance of the investor based on a predefined relationship between the balance return and the additional investment weight. An example of a method of calculating an investor-customized investment weight according to an investor's balance return is shown in FIG. 8 .

The present invention can also be implemented as computer-readable codes on a computer-readable recording medium. The computer-readable recording medium includes all types of recording devices in which data readable by a computer system is stored. Examples of computer-readable recording media include ROM, RAM, CD-ROM, magnetic tape, floppy disk, and optical data storage device. In addition, the computer-readable recording medium is distributed in a network-connected computer system so that the computer-readable code can be stored and executed in a distributed manner.

So far, the present invention has been looked at with respect to preferred embodiments thereof. Those of ordinary skill in the art to which the present invention pertains will understand that the present invention can be implemented in a modified form without departing from the essential characteristics of the present invention. Therefore, the disclosed embodiments are to be considered in an illustrative rather than a restrictive sense. The scope of the present invention is indicated in the claims rather than the foregoing description, and all differences within the scope equivalent thereto should be construed as being included in the present invention.

Claims

training a predictive model that outputs a predictive price using time series data of financial information;

calculating a plurality of predictive values by repeatedly inputting the time series data at the current time into the learned predictive model a predefined number of times;

identifying a distribution of the plurality of predictors; and

Calculating the investment weight based on the ratio of the first area of the upper distribution and the second area of the lower distribution determined based on the current price in the distribution;
The method of claim 1, wherein the training of the predictive model comprises:

generating augmented data based on the time series data; and

Including; using the time series data and the augmented data as training data of the predictive model;

The step of generating the augmented data comprises:

generating a network graph composed of at least one moving average line based on the time series data;

determining the distance between the value of each time point of the time series data and the moving average line in the mesh graph and the slope of the moving average line at each time point; and

An investment advisory method comprising a; generating augmented data including the distance and the inclination.
The method of claim 1, wherein calculating the investment weight comprises:

If the first area is larger than the second area, a positive investment ratio is calculated in proportion to the ratio of the first area to the second area, and if the first area is smaller than the second area, the first area and Calculating a negative investment weight proportional to the ratio of the second area; Investment advisory method comprising: a.
The method of claim 1,

The investment advisory method further comprising; correcting the additional investment weight corresponding to the return on balance of the investor based on a predefined relationship between the balance return and the additional investment weight by reflecting the investment weight.
a learning unit for learning a predictive model that outputs a predictive price using time series data of financial information;

a predictor for calculating a plurality of predictive values by repeatedly inputting the time series data at the current time into the predictive model a predefined number of times;

a distribution grasping unit which grasps a distribution of the plurality of predictors; and

Investment advisory apparatus comprising: an investment weighting calculator that calculates the investment weight based on the ratio of the first area of the upper distribution and the second area of the lower distribution determined based on the current price in the distribution.
6. The method of claim 5,

After generating a network graph composed of at least one moving average line based on the time series data, and determining the distance between the value of each time point of the time series data and the moving average line and the slope of the moving average line at each time point in the mesh graph, the time series The investment advisory apparatus further comprising a; a learning data generation unit for generating learning data including data, the distance, and the slope.
6. The method of claim 5,

Investment advisory apparatus further comprising a; a correction unit for correcting the additional investment weight corresponding to the return on balance of the investor based on a predefined relationship between the balance return and the additional investment weight by reflecting the investment weight.
A computer-readable recording medium in which a program for performing the method according to claim 1 is recorded.