KR102614595B1 - A stock price prediction system based on real-time macro index prediction - Google Patents

Abstract

Analyze the correlation between multiple macro indicators for stock prices, calculate the difference between the predicted indicator value and the actual value for each multiple macro indicator, and weight the macro indicator by reflecting the correlation in the difference between each macro indicator. It relates to a stock price prediction system based on real-time macro indicator prediction that calculates and reflects the calculated weights in predicted stock prices, comprising: a data collection unit that collects stock price information and economic indicators as time series data; a data normalization unit that calculates increments and normalizes the time series data; A correlation analysis unit that calculates the correlation coefficient between stock prices and each economic indicator; Neural network management department that trains LSTM neural networks for each stock price and economic indicator; A stock price and index prediction unit that predicts each stock price and economic index using a learned LSTM neural network; And, a configuration is provided that includes a stock price correction unit that obtains correction values for each economic indicator and corrects them by reflecting the obtained correction values in the predicted stock price.
By using the above system, the correlation between macro indicators for stock prices is analyzed, macro indicators are predicted in real time, and the influence of the macro indicators is weighted on the predicted stock prices, thereby predicting the time series of stock prices and simultaneously predicting the macroeconomic environment. By reflecting all of this, you can predict the stock price more accurately.

Description

Stock price prediction system based on real-time macro index prediction { A stock price prediction system based on real-time macro index prediction }

The present invention analyzes the correlation between a plurality of macro indicators for stock prices, calculates the difference between the predicted indicator value and the actual value for each of the plurality of macro indicators, and reflects the correlation in the difference between each macro indicator. This relates to a stock price prediction system based on real-time macro indicator prediction that calculates the weight of the indicator and reflects the calculated weight in the predicted stock price.

Recently, the global economy has stagnated due to infectious diseases such as coronavirus, and most countries, including the U.S. Federal Reserve System, have introduced policies to stimulate the economy by increasing the money supply. As the market was overflowing with funds, liquid funds flowed into the stock market, and domestic stock markets, including those in the U.S., rose sharply, leading to overheating of the stock market.

However, most stock investors tend to invest only by listening to recommendations from famous YouTubers or acquaintances without analyzing the company's financial statements, increasing the possibility of stock investment losses.

In general, stock price analysis largely uses fundamental analysis and technical analysis. Basic analysis is analyzing and predicting the value of a company through the company's financial statements, industry trends, CEO ability, etc. In this study, it means predicting the stock price through technical analysis. Technical analysis is basically the market price of the stock price used, Candle Chart, Moving Average, RSI, (Relative Strength Index), CCI (Commodity Channel Index), MACD (Moving Average Convergence and Divergence) along with basic stock price data such as high, low, closing price, and trading volume. ), Disparity, SIGNAL, and stochastic1, etc. are used [10-12]. However, this analysis is not only difficult, but also difficult for the general public to analyze, and even with this analysis, it is not easy to predict stock prices.

To solve these problems, technologies for predicting stock prices are being developed. As an example, methods such as analyzing past data and making inferences through time series [Patent Document 1], or using artificial intelligence to learn stock price fluctuation patterns and predict the future [Patent Document 2], etc. have been proposed.

However, there are various factors that affect stock prices, and it is virtually impossible to predict the future using past data. Additionally, the stock market does not repeat past trading patterns and is affected by global economic conditions such as overseas markets and economic conditions. In addition, it is affected by new situations at every moment, such as international relations and specific events. Therefore, even if a typical deep learning algorithm is applied, it is almost impossible to predict the constantly changing prices of stocks.

Therefore, it is necessary to predict stock prices by reflecting not only time series analysis of stock prices but also economic macro indicators and market conditions based on news.

Korean Patent Publication No. 10-1508361 (announced on April 8, 2015) Korean Patent Publication No. 10-1458004 (announced on November 4, 2014)

Lee Jong-girl. (2008). A comparative analysis of major macro indicators in five Latin American countries. Latin American and Caribbean Studies 26(2), 83-106 Sukha Shin. (2013). Analysis on the Relation between Foreign Investors in Korean Stock Markets and Macroeconomic Variables. INTERNATIONAL BUSINESS REVIEW 17(2), 89-107 Mahdy, F. E. INDUSTRIES STOCK RETURN REACTIONS TO RISK FACTORS: AN EMPIRICAL INVESTIGATION ON THE G-7 COUNTRIES. Guru-Gharan, K. K., Rahman, M., & Parayitam, S. (2009). Influences of selected macroeconomic variables on US stock market returns and their predictability over varying time horizons. Academy of Accounting and Financial Studies Journal, 13(1), 13. Qian, Y., Fan, Y., Hu, W., & Soong, F. K. (2014, May). On the training aspects of deep neural network (DNN) for parametric TTS synthesis. In 2014 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP) (pp. 3829-3833). IEEE. Mao, J., Xu, W., Yang, Y., Wang, J., Huang, Z., & Yuille, A. (2014). Deep captioning with multimodal recurrent neural networks (m-rnn). arXiv preprint arXiv:1412.6632. Pascanu, R., Gulcehre, C., Cho, K., & Bengio, Y. (2013). How to construct deep recurrent neural networks. arXiv preprint arXiv:1312.6026. Goodfellow, I., Bengio, Y., Courville, A., & Bengio, Y. (2016). Deep learning (Vol. 1, No. 2). Cambridge: MIT press. Medsker, L. R., & Jain, L. C. (2001). Recurrent neural networks. Design and Applications, 5. Selvin, S., Vinayakumar, R., Gopalakrishnan, E. A., Menon, V. K., & Soman, K. P. (2017, September). Stock price prediction using LSTM, RNN and CNN-sliding window model. In 2017 international conference on advances in computing, communications and informatics (icacci) (pp. 1643-1647). IEEE. Dong-Ha Shin, Kwang-Ho Choi, and Chang-Bok Kim. (2017). Deep Learning Model for Prediction Rate Improvement of Stock Price Using RNN and LSTM. Korean Institute of Information Technology, 15(10), 9-16. Hsieh, T., Hsiao, H., & Yeh, W. (2011). Forecasting stock markets using wavelet transforms and recurrent neural networks: An integrated system based on artificial bee colony algorithm. Applied Soft Computing, 11(2), 2510-2525. doi:10.1016/j.asoc.2010.09.007 Mahmud, M. S., & Meesad, P. (2014). Time series stock price prediction using recurrent error based neuro-fuzzy system with momentum. 2014 International Electrical Engineering Congress (iEECON). doi:10.1109/ieecon.2014.6925866

The purpose of the present invention is to solve the problems described above, using a LSTM (long short memory networks) neural network to learn using past data on stock prices, predicting stock prices with the learned neural network, and predicting stock prices and macro indicators. It provides a stock price prediction system based on real-time macro indicator prediction that adds weight to the predicted stock price by analyzing real-time macro indicators through correlation.

In addition, the purpose of the present invention is to analyze the correlation between a plurality of macro indicators for stock prices, calculate the difference between the predicted indicator value and the actual value for each of the plurality of macro indicators, and calculate the correlation between the differences between each macro indicator. It provides a stock price prediction system based on real-time macro indicator prediction that calculates the weight of the macro indicator and reflects the calculated weight in the predicted stock price.

In other words, the purpose of the present invention is to analyze and predict macro indicators that affect stock prices by utilizing artificial intelligence deep learning techniques developed under existing automatic trading conditions, and to apply weight to individual stock price predictions through correlations that affect stock prices. It provides a stock price prediction system based on real-time macro indicator prediction that predicts stock prices.

In addition, the purpose of the present invention is to provide a stock price prediction system based on real-time macro indicator prediction that reflects weights on stock prices predicted by artificial intelligence through stock market news text mining because stock prices react sensitively to real-time stock market news.

In order to achieve the above object, the present invention relates to a stock price prediction system based on real-time macro indicator prediction, which includes a data collection unit that collects stock price information and economic indicators as time series data; a data normalization unit that calculates increments and normalizes the time series data; A correlation analysis unit that calculates the correlation coefficient between stock prices and each economic indicator; Neural network management department that trains LSTM neural networks for each stock price and economic indicator; A stock price and index prediction unit that predicts each stock price and economic index using a learned LSTM neural network; And, it is characterized by including a stock price correction unit that calculates the correction value of each economic indicator and corrects it by reflecting the obtained correction value in the predicted stock price.

In addition, the present invention is a stock price prediction system based on real-time macro indicator prediction, in which the system, the neural network learning unit, collects news data of the stock item and performs text mining to calculate the news index of the stock price. It further includes a unit, wherein the stock price correction unit multiplies the predicted stock price by a news index and then adds a correction value of the economic indicator to the multiplication result to estimate the final stock price.

In addition, the present invention is a stock price prediction system based on real-time macro indicator prediction, wherein the neural network management unit provides total normalized incremental time series data M(-N+1), M(-N+2), ....,M A number of batch sets are sequentially generated from (0), and one set is formed with a batch size of n+1 to form a series of N-n batch sets, and each batch set includes n historical data sets and the corresponding set date. It is characterized by generating learning data by configuring the correct answer of the prediction data at the time.

In addition, the present invention is a stock price prediction system based on real-time macro indicator prediction, wherein the stock price correction unit multiplies the actual value of each indicator by subtracting the predicted value by 100, and adds a correlation coefficient (correlation coefficient between each indicator and the stock price) to the multiplied value. By multiplying, the correction value w _i for each index of the relevant index is calculated, and the correction value w _i for each index is averaged.

In addition, the present invention is characterized in that, in a stock price prediction system based on real-time macro indicator prediction, the stock price is used as the midpoint between the high and low prices of the stock price.

In addition, the present invention is a stock price prediction system based on real-time macro indicator prediction, wherein the neural network is a long short-term memory networks (LSTM) neural network.

As described above, according to the stock price prediction system based on real-time macro indicator prediction according to the present invention, by analyzing the correlation of macro indicators with stock prices, predicting macro indicators in real time, and weighting the influence of macro indicators on the predicted stock price, The effect of predicting stock prices more accurately is achieved by reflecting the macroeconomic environment at the same time as predicting the time series of stock prices.

1 is an exemplary configuration diagram of an overall system for implementing the present invention.
Figure 2 is a block diagram of the configuration of a stock price prediction system based on real-time macro indicator prediction according to an embodiment of the present invention.
Figure 3 is a table illustrating time series data or incremental time series data of each economic indicator according to an embodiment of the present invention.
Figure 4 is a table showing the correlation or correlation coefficient between each economic indicator and stock prices according to an embodiment of the present invention.
Figure 5 is a diagram illustrating the structure of an LSTM neural network according to an embodiment of the present invention.
Figure 6 is programming code for a method for learning and predicting stock prices and economic indicators according to an embodiment of the present invention.
Figure 7 is a graph showing the results of stock price prediction analysis according to an embodiment of the present invention.
Figure 8 is a table illustrating result values in the process of calculating correction values for each economic indicator according to an embodiment of the present invention.

Hereinafter, specific details for implementing the present invention will be described with reference to the drawings.

In addition, in explaining the present invention, like parts are given the same reference numerals, and repeated description thereof is omitted.

First, the configuration of the entire system for implementing the present invention will be described with reference to FIG. 1.

As shown in Figure 1, the entire system for implementing the present invention includes a user terminal 10, a stock price prediction server 30 that predicts stock prices, a stock price information server 50 that provides stock price information, and macro indicator information. It consists of an economic information server 60 that provides news, etc. Additionally, it may further include a database 40 that stores necessary data.

First, the user terminal 10 is a terminal used by a user, and is a general computer terminal or dedicated terminal with a computing function, such as a smartphone, tablet PC, laptop, or personal computer (PC).

The user terminal 10 can access the stock price prediction server 30 and receive a stock price prediction service provided by the stock price prediction server 30. That is, the user terminal 10 inputs an item from the stock price prediction server 30, and requests and receives the expected stock price for the input item.

Preferably, the expected stock price consists of the highest price target and the lowest price target.

Next, the stock price information server 50 is a typical server that provides stock price information, and is a server operated by a securities company, stock price information provider, etc. There may be a plurality of stock price information servers 50, and each stock price information server 50 may provide different types of stock price information.

Preferably, the stock price information consists of high price, low price, closing price, etc. Additionally, each stock price information consists of date-specific information (data by date). In other words, the stock price information includes information on stock prices traded on the current or past trading business days.

Next, the economic information server 60 is a server that provides information related to the economy, such as economic macro indicators and news, and is operated by each organization, newspaper company, broadcasting company, or public announcement agency. The economic information server 60 provides economic indicators, news, etc. online.

Next, the stock price prediction server 30, which is a normal server, predicts stock prices according to a request from the user or the user terminal 10 and provides predicted stock price information to the user. Preferably, a specific stock name is input, the stock price for the stock is predicted, and predicted stock price data for the stock is provided.

In particular, the stock price prediction server 30 is equipped with an LSTM neural network and trains the LSTM neural network using past stock price information. Then, the current stock price information is input into the trained LSTM neural network to obtain predicted stock price data.

Additionally, the stock price prediction server 30 can provide predicted stock price data with the highest and lowest target prices. At this time, preferably, the highest and lowest target prices are extracted by setting an error range in the predicted stock price.

Additionally, the stock price prediction server 30 receives stock price information from the stock price information server 50 and utilizes it.

Additionally, the stock price prediction server 30 is equipped with a neural network and trains the neural network using past macro indicator information. Then, the current macro indicator is input into the trained neural network to obtain the predicted macro indicator.

Additionally, the stock price prediction server 30 is equipped with a text miner and performs mining analysis on news using the text miner. Through a text miner, a number is extracted to determine whether the current news is positive or negative (a number for good news or bad news). This figure is called the news rating.

In particular, by entering the stock name into the text miner, the news rating for the stock is extracted. Additionally, the text miner analyzes news data from the news server 60 and extracts news ratings.

In addition, the stock price prediction server 30 calculates weights based on predicted macro indicators, or converts news ratings into weights, and calculates final predicted stock price data by adding the corresponding weights to the previously obtained predicted stock price data.

Next, the database 40 includes a stock price information DB 41 that stores stock price information, an economic indicator DB 42 that stores economic macro indicators, and a correlation information DB that stores the results of analyzing the correlation between stock prices and macro indicators. It consists of (43). However, the configuration of the database 40 is only a preferred embodiment, and when developing a specific device, it may be configured in a different structure according to database construction theory, taking into account ease of access and search and efficiency.

Next, the configuration of the real-time stock price prediction system 30 according to an embodiment of the present invention will be described with reference to FIG. 2.

The real-time stock price prediction system 30 according to the present invention may be implemented in the form of a stock price prediction server 30. Alternatively, the real-time stock price prediction system 30 may be implemented in a server-client form, and a client module may be installed in the user terminal 10.

As shown in Figure 2, the stock price real-time prediction system 30 according to an embodiment of the present invention includes a data collection unit 31 that collects stock price-related data such as stock price information, economic macro indicators, news, stock prices, macro indicators, etc. Data normalization unit 32, which calculates and normalizes increments to the value, correlation analysis unit 33, which analyzes the correlation between stock prices and macro indicators, neural network management unit 34, which configures and learns a neural network, stock price and macro indicators It consists of a stock price and index prediction unit (35) that predicts indicators, and a stock price correction unit (37) that corrects the predicted stock price. Additionally, it further includes a news analysis unit 37 that analyzes news.

First, the data collection unit 31 collects stock price-related data such as stock price information, economic macro indicators, and news.

Stock price information consists of high price, low price, closing price, etc. Additionally, each stock price information consists of date-specific information (data by date). In other words, the stock price information includes information on stock prices traded on the current or past trading business days.

In addition, macro indicators are economic indicators that aggregate the economic situation of a specific country and indicate the economic situation of that country. Macro indicators consist of exchange rate, interest rate, money supply, GDP, GDP growth rate, unemployment rate, inflation rate, current account balance, consumer index, consumer survey, production index, number of new housing supply, number of unemployment claims, long-term government bond interest rate, raw materials, etc. [Non-patent Document 1]. In addition to macro indicators, it can be applied to all economic indicators. For example, instead of macro indicators, interest rates, exchange rates, futures indices, etc. can be collected on a daily basis. Additionally, historical data can be downloaded and used on a monthly basis from the fred.stlouisfed.org site.

Additionally, macro indicators are composed of period-specific information (period-specific data). Preferably, each macro indicator consists of monthly data. In other words, the date unit is set to month. If the economic indicator is data that changes on a daily basis, it can be set on a daily basis. Below, for convenience of explanation, it is explained in terms of days (or date units), but the date unit can be set to normal period units such as days, months, or weeks.

Data collected quarterly, such as GDP, are divided by month and converted to monthly data. The usual method of dividing quarterly data is applied. For example, quarterly data can be divided by 3 and used as data for each month, or interpolation can be applied to convert each month's data to have continuity according to the values of the previous quarter and the next quarter.

Additionally, news information is news data written in text. News can be classified depending on the provider. In other words, the news provider is the Financial Supervisory Service, each newspaper or broadcasting company, etc.

Meanwhile, if the actual value has not yet been announced in the relevant period, the consensus value (or estimated value) in the market can be collected and replaced with the corresponding value.

Next, the data normalization unit 32 calculates increments for values such as stock prices and macro indicators within a specific period and normalizes them.

Stock prices and each macro indicator have different units. Therefore, the increment is calculated for the value of each stock price or macro indicator, and the increment value is used instead of the corresponding value. Then, set (reset) the increment value of the standard day to 0. For example, if data was collected from January 2010 to December 2020, the value of the data for January 2010 is set to 0.

The increment of each data (stock price, macro indicator, etc.) is calculated by subtracting the value of the previous day (previous month) from the value of the current day (month) and dividing it by the value of the previous day (previous month). In other words, the increment △M(t) of day t (month) is obtained by the following equation.

[Equation 1]

Here, M(t) represents the data value at time t of the corresponding data.

Each incremental data is time series data. If the size of the entire time series data is N, the time series data of each stock price or macro indicator is data by a series of dates, such as M(-N+1), M(-N+2), ...., M( 0) is. M(0) is the data of the current day (month), S(-k) is the incremental data before the k day (month), and M(0) is the incremental data of the current day (month).

Also, preferably, the stock price data is used as the median value of the stock price (the middle value between the high and low prices). In other words, the median value is obtained as the average of the high and low prices as shown in the following equation.

[Equation 2]

Here, t is the date, M(t) is the median value on day t, and H(t) and L(t) represent the high and low prices on day t, respectively.

Next, the correlation analysis unit 33 analyzes the correlation between stock prices and each macro indicator.

That is, the correlation analysis unit 33 performs a correlation analysis on the incremental values of each stock price and macro index within a specific period to obtain the correlation coefficient between the stock price and the macro index. In other words, the correlation coefficient is used as the analysis result of correlation.

There is a lot of research being done on the relationship between macro indicators and stock price fluctuation rates. Although there are some differences, nominal economic variables such as money supply and prices usually affect stock price fluctuations, but other real economic indicators are analyzed to have no significant impact. There is [Non-patent Document 2-4].

There are a variety of stock items, and it is believed that more accurate predictions can be made if individual stock items that have a high correlation with macro indicators are analyzed and the corresponding macro indicators are predicted for reference when predicting individual stock items.

As shown in the table of Figure 3, the values of macro indicators use incremental values. In other words, since the units of macro indicators are different, the previously obtained incremental values are used for comparison and correlation analysis.

The table in Figure 4 analyzes the incremental values of macro indicator data for 10 years from 2011 to 2021, and shows the correlation between macro indicators and stock price items (NASDAQ index NASDAQ, Dow index DOW) as an index (or correlation coefficient). there is. At this time, both stock prices and macro indicators use increments. The correlation coefficient is obtained by normal correlation analysis.

The closer the index or correlation coefficient is to 1, the stronger the correlation, and the closer it is to 0, the absence of correlation, meaning there is a strong inverse correlation.

Preferably, only indicators whose correlation coefficient is greater than or equal to a predetermined threshold (or threshold coefficient) are selected and used.

The table in Figure 4 shows only those with a correlation index of 0.5 or higher among various macro indicators. In the table in Figure 3, since GDP is compiled on a quarterly basis, there is no data after October 2020, and in order to unify monthly data, quarterly incremental values are calculated at an equal ratio and expressed monthly.

The larger the number of unemployment claims, the more it affects the decline in stock prices, showing an inverse correlation. Therefore, the value of the correlation coefficient indicating correlation has a minus (-) value.

Next, the neural network management unit 34 configures a long short memory networks (LSTM) neural network for each stock price and macro indicator and trains the LSTM neural network.

Artificial intelligence is defined as a computer program that artificially implements human abilities such as perception, reasoning, and learning, or a computing system including these [Non-patent Document 7]. Artificial intelligence is a general term, and machine learning is a field of artificial intelligence that learns on its own without human intervention through repetitive learning. Deep learning is a further development of machine learning and makes more accurate predictions with multiple hidden layers. This is a way to do this. Representative deep learning networks include ANN (Artificial Neural Network), DNN (Deep Neural Network), RNN (Recurrent Neural Networks), and LSTM (Long Short Term Memory) [Non-patent Document 4-6]. Deep learning networks are applied to make predictions based on past data.

RNN and LSTM are easy to analyze sequential and repetitive data and are widely used in stock price analysis [Non-patent Document 8]. Predictions are made by continuously repeating the influence of past data on the present through weights. Therefore, RNN and LSTM are widely used in stock price prediction [Non-patent Document 9].

In the present invention, rather than predicting stock prices using only past data, each macro indicator (or economic indicator) is analyzed and predicted, and the weight value is adjusted according to the difference compared to the actual stock price to best match the actual stock price. Predict stock prices through repeated learning. In addition, the weight value is adjusted according to the correlation between macro indicators to make optimal predictions, and the weight value is also adjusted through iterative learning.

In particular, LSTM (Long Short Term Memory networks) neural network is a type of RNN (Recurrent Neural Network) and has a structure in which past data can affect the future. Therefore, the LSTM neural network is the most suitable model for stock price prediction.

In the existing RNN neural network, the values between [-1, 1] are continuously multiplied due to the chain rule, which is a long-term dependency problem, so the value gets smaller as it goes forward and eventually disappears. Therefore, the existing RNN method has a problem in that parameters are not updated. LSTM is a proposed method to solve this problem.

The existing RNN simply calculated St=tanh(Uxt+WSt) when calculating the hidden state (St) in the hidden layer, but in LSTM, there are a total of four calculation processes.

As shown in Figure 5, there are four neural network layers in the hidden layer of LSTM. In other words, it can be confirmed that there are four neural network layers within the central square shape of Figure 5. Additionally, the core of LSTM is that a very minor linear operation is performed on the horizontal line at the top of the neural (A), and the information is passed on to the next step without much change through the entire chain.

Meanwhile, an LSTM neural network is constructed to predict stock prices and each economic indicator (or macro indicator). That is, it consists of an LSTM neural network for stock prices, an LSTM neural network for the consumer index, and an LSTM neural network for the production index.

In particular, for each time series data, it is trained with historical data for a specific period (e.g., 30 days, hereinafter referred to as the reflection period), and then predicted for one day and compared with actual data. Through this comparison, the training set is learned by moving into the future in units of reflection periods (30 days, for example).

Also, preferably, the LSTM neural network for stock prices is learned using the high and low prices of stock prices as default values. Additionally, the LSTM neural network is trained on the stock price of one stock. Therefore, in order to predict the stock price of a new stock (or other stock), the LSTM neural network is initialized and retrained with the stock's past stock price data. Or, as another embodiment, the LSTM neural network includes an LSTM neural network corresponding to each item to be predicted. In other words, it has multiple LSTM neural networks corresponding to each stock, and learns each stock from its past stock price data.

Also, preferably, the LSTM neural network is learned for one macro indicator (or economic indicator). Therefore, to predict other economic indicators, the LSTM neural network is initialized and retrained with past data of the relevant economic indicator. Or, as another embodiment, the LSTM neural network is equipped with an LSTM neural network corresponding to each economic indicator to be predicted for each indicator. In other words, a number of LSTM neural networks corresponding to each indicator are provided, and each indicator is learned from the past time series data of that indicator.

Meanwhile, the neural network management unit 34 trains the corresponding neural network with incremental time series data of each stock price or each economic indicator (or macro indicator).

Let the size of the entire time series data be N, and set the batch size to n. At this time, N > n, and preferably, N and n are set to satisfy N > 2n. In other words, the purpose is to make the number of batches larger than the batch size.

The full incremental time series data is data by a series of dates, such as M(-N+1), M(-N+2), ...., M(0). M(-k) is the normalized incremental data k days ago, and M(0) is the normalized incremental data of the current day.

At this time, a batch is created with batch size n. If the total data size is N, the following N-n batches are created.

[ M(-N+1), M(-N+2), M(-N+3), ..., M(-N+n) ], M(-N+n+1)

[ M(-N+2), M(-N+3), M(-N+4), ..., M(-N+n+1) ], M(-N+n+2)

...

[ M(-n-1), M(-n), M(-n+1), ..., M(-2) ], M(-1)

[ M(-n), M(-n+1), M(-n+2), ..., M(-1) ], M(0)

As above, N consecutive time series data (incremental data) are formed into a set with a batch size of n+1, and the sets are formed sequentially to form a series of N-n batch sets. At this time, the last data of each batch set is set as the result value (real value) of that set. In other words, each batch consists of n historical stock price data sets and the next day's stock price (the true stock price to predict) (the correct stock price) for that set.

Then, the neural network management unit 34 uses each batch set as one learning data and trains the LSTM neural network. After learning, stock investors can decide whether to sell or buy stocks by predicting the price for the next day.

In particular, the neural network management unit 34 normalizes the data set (or batch set) to predict stock or economic indicators. Preferably, the training data and the test data set that validates the training data are divided in a 9:1 ratio. That is, the training set is set to 90% of the data set, and the test set is set to 10% of the data set. These ratios may vary depending on stocks or economic indicators predicted in the future.

Pseudocode for the specific learning method is shown in Figure 6.

In Figure 6, part ① used pandas to use csv files, numpy for matrix operations, matplotlib.pyplot for data visualization, and Keras to create a deep learning model. ③ The window size is analyzed in units of 20, and the meaning of 20 is to predict one data after analyzing 20 pieces of data. ④ Using the function called Sequential(), set 90% of the 20 window sizes as the training set, and set the remaining 10% as the test set.

In Figure 6, the NASDAQ stock index prediction is divided into a training set and a test set for learning.

Next, the stock price and index prediction unit 35 inputs the latest data set into the learned LSTM neural network and obtains a predicted value of the data.

The most recent data set (or batch) is as follows:

[ M(-n+1), M(-n+2), ..., M(-1), M(0) ]

That is, the recent data set consists of the latest time series data as many as the number of batches, including the most recent date or the current day's date.

When the latest data set is input to the LSTM neural network, the LSTM neural network outputs the predicted value of the corresponding data, that is, the predicted value M(1) for the next day (next month). The output M(1) is the predicted value of the data.

The stock price and index prediction unit 35 obtains predicted values for each stock price and each economic index.

Figure 7 shows actual values and predicted result values for stock prices. In Figure 7, red represents the actual stock price data, and blue represents the predicted result value.

Next, the news analysis unit 37 collects news data for the relevant item, performs text mining, and calculates mood data for the relevant item.

Preferably, the news analysis unit 37 uses a commercially available tool such as YTextMiner. Text miners (such as YTextMiner) are text analysis tools that analyze real-time stock market news and determine whether news corresponding to stock items by keyword is good news or bad news.

Mood data is a numerical indicator of whether there is good or bad news for the stock. In case of good news, the value of mood data increases, and in case of bad news, the mood value decreases.

Preferably, the mood data (or mood value) has a value between (1-α) and (1+α). At this time, α has a value of 1% to 30%. Mood data is used as a news index.

Next, the stock price correction unit 37 corrects the predicted stock price using the correction value of the economic indicator (or macro indicator). Additionally, mood data based on news analysis is added to correct the predicted stock price.

The table in Figure 8 shows the process of calculating correction values based on economic indicators or macro indicators.

The method of calculating the correction value is to multiply the actual value of each indicator by subtracting the predicted value by 100, and multiply the multiplied value by the correlation coefficient (correlation coefficient between each indicator and the stock price) to calculate the correction value w _i for each indicator. And the final correction value is obtained by averaging them. Correction value w _i refers to the correction value (correction value for each indicator) for economic indicator i.

Meanwhile, the predicted value of each indicator is the value previously predicted through a neural network, and the actual value is the actual value of the indicator. However, if the actual value has not yet been announced, the consensus value (or estimated value) in the market is collected and replaced with the corresponding value.

Then, the stock price correction unit 37 multiplies the stock price forecast by the news mining index (or news index) and adds the index correction value to the value to calculate the final stock price (or corrected stock price). In other words, the final or corrected stock price is calculated as follows:

[Equation 3]

Adjusted stock price = stock price forecast × news index + index correction value

Here, the stock price forecast is the stock price forecast predicted by the LSTM neural network, and the news index is used as mood data previously obtained by the news analysis unit 37.

Therefore, through LSTM, predictions are made for individual stock prices, predictions are made for macro indicators that have a high or inverse correlation with individual stock prices, and then the predicted values of the macro indicators are reflected in the predicted stock prices. Through this, the accuracy of prediction can be improved compared to the existing method of analyzing only past stock price time series data.

In particular, by reflecting the difference between the actual and predicted values of economic indicators, stock prices can be predicted more accurately by reflecting the degree of surprise in economic indicators.

In addition, stock prices may fluctuate due to unpredictable stock market news or the words of celebrities, so if stock market news is judged as good or bad news for the stock price through real-time text mining and reflected in the stock price, the stock price prediction system that has been used so far is used. Accuracy can be further improved.

In other words, in the present invention, accurate stock price prediction can be made by learning technical analysis through LSTM, analyzing various macro indicators and correlations to adjust weight values, and then predicting the final stock price after text mining real-time stock price news. I hope there will be.

Although many technologies have been proposed to predict stock prices using deep learning algorithms of artificial intelligence, most existing stock price prediction algorithms analyze past time series data. Therefore, there are limitations because it cannot reflect unexpected situations in real time. However, if you use artificial intelligence machine learning techniques to learn repetitive patterns and correct prediction errors through real-time news mining, you can make the same level of judgment as a fund manager and trade stock prices yourself. The present invention does not predict stock prices by simply analyzing stock price indicators, but predicts stock prices by finding new changing patterns through real-time repetitive learning through macro indicators and news mining.

Above, the invention made by the present inventor has been specifically described based on examples, but the present invention is not limited to the examples and, of course, can be changed in various ways without departing from the gist of the invention.

10: User terminal 30: Stock price prediction server
31: data collection unit 32: data normalization unit
33: Correlation analysis unit 34: Neural network management unit
35: Stock price and index prediction department 36: News analysis department
37: Stock price correction unit
40: Database 50: Stock price information server
60: Economic information server

Claims (6)

In a stock price prediction system based on real-time macro indicator prediction,
A data collection unit that collects stock price information and economic indicators as time series data;
a data normalization unit that calculates increments and normalizes the time series data;
A correlation analysis unit that calculates the correlation coefficient between stock prices and each economic indicator;
Neural network management department that trains deep neural networks for each stock price and economic indicator;
A stock price and index prediction unit that predicts stock prices and economic indicators using a learned deep neural network; and,
It includes a stock price correction unit that calculates the correction value of each economic indicator and corrects it by reflecting the obtained correction value in the predicted stock price,
The correlation analysis unit performs correlation analysis on the incremental values of each stock price and economic index to calculate the correlation coefficient between the stock price and economic index,
The neural network management unit configures neural networks that predict each stock price and economic index, and trains each neural network with incremental time series data of each stock price or economic index,
The stock price correction unit multiplies the actual value of each economic indicator by subtracting the predicted value by 100, multiplies the multiplied value by the correlation coefficient (correlation coefficient between each indicator and the stock price), calculates the correction value w i for each indicator, and calculates the correction value w _i for each indicator. Correction is made by averaging the correction values w _i for each indicator,
A stock price prediction system based on real-time macro indicator prediction, wherein the stock price correction unit replaces the actual value with a market consensus value when the actual value is not announced.
According to paragraph 1,
The system further includes a news analysis unit, wherein the neural network management unit collects news data of an item of a corresponding stock price and performs text mining to calculate a news index of the corresponding stock price,
The stock price correction unit multiplies the predicted stock price by a news index and then adds a correction value of the economic indicator to the multiplication result to estimate the final stock price. A stock price prediction system based on real-time macro indicator prediction.
According to paragraph 1,
The neural network management unit sequentially generates a plurality of batch sets from the entire normalized incremental time series data M(-N+1), M(-N+2), ...., M(0), and the batch size is A set is formed with n+1 to form a series of Nn batch sets, and each batch set is composed of n past data sets and the correct answer of the predicted data for that set, thereby generating learning data. A stock price prediction system based on real-time macro indicator prediction.
delete According to paragraph 1,
A stock price prediction system based on real-time macro indicator prediction, characterized in that the stock price is used as the midpoint between the high and low prices of the stock price.
According to paragraph 1,
The deep neural network is a stock price prediction system based on real-time macro indicator prediction, characterized in that the LSTM (long short-term memory networks) neural network.

Images