TWI559249B - Computer implemented method and computer system for forecasting stock performance based on patent big data - Google Patents

Description

Patent big data forecasting stock selection method and computer system realized by computer

The invention relates to a digital content processing technology, in particular to a computer-based calculation based on a huge amount of patent data to explore the relevance of patent data to corporate financial information, and then to establish a patent-based stock selection method and computer system.

With the rapid development of the technology industry and the importance of intellectual property, patents have been regarded as an important indicator of industry or technology. For rights holders, patents are not only a defense tool to protect innovation and products, but also the best attack weapon to compete with competitors on the industrial stage. If a company has more patents and is more representative and influential in its competitive field, patents become an extremely important competitive information. The World Intellectual Property Organization (WIPO) has reported that 80% of the technical content of patents has not been disclosed in other papers, magazines, and encyclopedias. Patents are the specific manifestations of R&D innovation. Units with superior patent quantity and quality have more advantages in energy and quality of R&D innovation than other competitors. Because patents have legal exclusive rights, they have an oligopolistic effect on the market. Therefore, for technology-based enterprises based on technology research and development, the number and quality of patents are more advantageous, and their product sales and performance also have a considerable degree of superiority. In the prior art, many academic studies have pointed out that the quantity information of patents leads the sales information of products, and has a leading effect on explaining the development of the market. Therefore, the method of predicting market information with patent data has gradually become an important research topic in investment and corporate evaluation.

In the prior art, US Patent No. 6175824 and US6832171 first proposed to evaluate the financial performance of a listed company by using patent materials. U.S. Patent No. 6,177,824 and US Pat. No. 6,831,171 for listing The company's patents with a large number of patents analyze the correlation between the patent indicators of listed companies over the years and the financial indicators through the multiple regression analysis model. Finally, the evaluation equation based on patent indicators is derived, and the formula is used to calculate its scientific and technological value. Then compare the market-to-book ratio (Market-to-Book Ratio) with this innovative value. If the technology value is greater than the price-to-book ratio, it is considered to have investment potential. If the technology value is less than the price-to-book ratio, it is regarded as no investment value. A tool for investing in stock picking. Its patentee CHI Research has pushed this evaluation equation based on patent data to the public, and established its own proprietary business profit model. The prior art US6175824 and US6832171 have their own uniqueness, but the disadvantage is that the equation for analyzing the patent index and the financial index is a typical multiple regression analysis model, and the self-variable is a time-series patent indicator, since the variable is a financial indicator of the same time series. Therefore, the multivariate regression equation established, when inputting the patent index at a certain point in time, the output obtained is the corresponding financial indicator at that point in time, rather than the "future" financial indicator.

In the prior art, US patents US6556992, US7657476 and US7716226, based on the patent maintenance rate, combined with other patent indicators, developed a set of evaluation methods for the innovation ability of listed companies by multiple regression analysis models, and selected accordingly. 300 US listed companies as sample stocks, after weighting calculations, released the world's first patent index (OT300 Patent Index). However, the multivariate regression analysis models used by US6556992, US7657476 and US7716226 have the same time series of the self-variable patent index and the variable financial indicator. Although the self-variable is different from the prior art US6175824 and US6832171, the established multiple regression equation When inputting a patent indicator at a certain point in time, the output obtained is still the corresponding financial indicator at that point in time, rather than the "future" financial indicator.

We must understand the practical operation of investing in stock picking. When investing in an investment institution, it is not hoping to make a profit now, but hopes to make a profit at some point in the future. In other words, when investing in an investment institution, it is hoped that the current information will have the effect of “predicting” the profit of “future”, so as to reduce investment risks and ensure investment efficiency. The above-mentioned prior art US6175824, US6832171, US6556992, US7657476 and US7716226 are substantially incapable of "predicting" effects. On the other hand, the above prior art US6175824, US6832171, US6556992, US7657476 and US7716226 are all based on the United States. The model and information of the authorized patent materials are modeled. The core patent indicators used by US6175824 and US6832171 are the citations of patent citations and non-patent literature used in the Novelty review and the Creative/Nonobviousness review, including the former citation and the post citation ( Forward Citation); The most patented index used by US6556992, US7657476 and US7716226 is the maintenance rate of invention patents. However, there is no utility model system in the US patents. However, for the vast majority of patent materials in mainland China, there are more utility model patents that do not require substantive examination procedures, and fewer invention patents. For invention patents, the authorized patents passed through entity examination are more Less, more early public patents. In addition, for the invention of patents granted, the citation information published by the patent database is gradually disclosed in recent years, and the patents have not been fully traced back to the previous authorized patents; The citation, which is the comparison file used in the examination of the patent entity (PE), does not disclose the post citation, that is, other patents that are examined as a comparison file after the patent is granted. Therefore, the contents and methods disclosed in the prior art US Pat. No. 6,176, 824, US Pat. No. 6,823, 721, and US Pat. No. 6,556, 992, and US Pat. No. 7,716,226, do not apply to the patent information of the Chinese mainland, and therefore cannot evaluate the financial information of the listed companies in mainland China.

Mainland China patent CN201410283508.7 and Republic of China patent TW104105237 first disclosed a financial indicator prediction model and computer system for patent data and financial data construction of listed companies in mainland China. Based on this prior art, the present invention provides further optimization and improvement.

Based on the improvement of the prior art described above, the present invention provides a patent big data forecasting stock selection method and a computer system realized by a computer, which not only contributes to the development of the technical strength of patent data analysis and utilization, but also promotes the positive investment method in the investment field. Development, and positive support for the research and development and innovation of industrial technology.

First, the present invention proposes a patent big data prediction stock selection method (100) implemented by a computer, comprising the following steps: (110) setting parameters: multiple patent entities (PE), one time length (T0), one time lead period (L), a first time period (T1), a second time period (T2), used to describe each patent a plurality of patent indicators (PI) within a length of time (T0) and at least one financial indicator (FI), wherein the first time period (T1) and the second time period (T2) have the same length of time (T0), the end date of the second time period (T2) (T20) is one time behind the end date (T10) of the first time period (T1) (L); (120) Collecting data: each patent entity (PE) During the first time period (T1), the patent index (PI) and the financial indicator (FI) correspond to a plurality of first patent indicator data (121P) and a plurality of first financial indicator data (121F), collected Each patent entity (PE) has a plurality of second patent indicator data (122P) and a plurality of second financial indicator data (122F) corresponding to the patent indicator (PI) and the financial indicator (FI) in the second time period (T2). (130) forming first panel data (121P), first financial indicator data (121F), second patent indicator data (122P), and second financial indicator data (122F) into a first panel data (131) (140) providing a normal distribution conversion program (141) for converting the first panel data (131) into a second panel data (142); (150) providing a first based on the time lead (L) Time An inter-sequence operation program (151), at least one first fitting coefficient (152), and a first threshold value (153) corresponding to the first fitting coefficient (152), the first time series operation program (151) The variable is a patent indicator (PI) of the second panel data (142), the variable is the financial indicator (FI) of the second panel data (142); (160) the first time series operation program (151) is used successively, and the operation is performed. The second panel data (142) is selected from a plurality of patent indicators (PI) to obtain a plurality of patent core indicators (PCI) that the first fitting coefficient (152) meets the first threshold (153); (170) is provided based on a second time series operation program (171) of the time lead period (L), at least one second fitting coefficient (172), and a second threshold value (173) corresponding to the second fitting coefficient (172), The independent variable of the second time series operation program (171) is all the patent core indicators (PCI) in the second panel data (142), and the variable is the financial indicator (FI) of the second panel data (142); (180) The second time series operation program (171), the calculation generates a patent leading equation (181), and the patent leading equation (181) is composed of a plurality of patent leading indicators (P). LI) and the corresponding weighting factor (182) of each patent leading indicator (PLI). The patent leading indicator (PLI) is selected by the patent core indicator (PCI). In the patent leading equation (181), the second fitting coefficient (172) of each patent leading indicator (PLI) meets the second threshold (173); (190) the first patent indicator data (121P) is introduced into the patent. The leading equation (181) generates a patent lead score (191) for each patent entity (PE); and (200) selects shares based on the patent lead score (191).

A further object of the present invention is to provide a patented big data predictive stock selection method (600) implemented by a computer, comprising the following steps: (610) setting parameters: multiple patent entities (PE), one time lead (L), one First year (T1), one second year (T2), multiple patent indicators (PI) and a finance used to describe each patent entity (PE) in the first year (T1) and the second year (T2) Indicators (FI), patent indicators (PI) include indicators describing invention patents, indicators describing invention patents, indicators describing utility model patents, indicators describing design patents, and indicators describing effective patents, among which patent entities (PE) is a listed company, and the financial indicator (FI) is the reinstatement closing price (SP) of the last trading day of each patent entity (PE) in the first year (T1) and the second year (T2), the first year The length of time between (T1) and the second year (T2) is one year, and the end date of the second year (T2) (T20) is one time behind the end of the first year (T1) (T10). L), time lead (L) is a quarter; (620) data collection: each patent entity (PE) in the first year (T1) Within, the first patent indicator (PI) and the financial indicator (FI) correspond to a plurality of first patent indicator data (121P) and a plurality of first financial indicator data (121F), collecting each patent entity (PE) in the second year ( Within T2), a plurality of second patent indicator data (122P) corresponding to the patent indicator (PI) and the financial indicator (FI) and a plurality of second financial indicator data (122F); (630) the first patent indicator data ( 121P), first financial indicator data (121F), second patent indicator data (122P), and second financial indicator data (122F) constitute a first panel data (131); (640) provide a normal distribution conversion program (141) converting first panel data (131) into a second panel data (142); (650) providing a one-time Granger causality test model based on time lead (L) (151), at least one first fitting coefficient (152), and a first threshold (153) corresponding to the first fitting coefficient (152), and the independent variable of the unary Granger causality test model (151) is A patent indicator (PI) of the second panel data (142), the variable is the financial indicator (FI) of the second panel data (142); (660) successively uses the one-dimensional Granger causality test model (151), the second operation The panel data (142) selects a plurality of patent core indicators (PCI) from a plurality of patent indicators (PI), and the first fitting coefficient (152) of each patent core indicator (PCI) meets the first threshold (153). (670) providing a time-leading (L)-based multivariate Granger causality test model (171), at least one second fit coefficient (172), and a second corresponding to the second fit coefficient (172) The threshold (173), the independent variable of the multivariate Granger causality test model (171) is all the patent core indicators (PCI) in the second panel data (142), and the variable is the financial indicator of the second panel data (142) ( FI); (680) through the multivariate Granger causality test model (171), the calculus generates a patent-leading equation (181), and the patent-leading equation (181) consists of multiple The leading edge indicator (PLI) and the corresponding patent leading indicator (PLI) corresponding weight coefficient (182), the patent leading indicator (PLI) selected by the patent core indicator (PCI), in the patent leading equation (181), each The second fitting coefficient (172) of the patent leading indicator (PLI) meets the second threshold (173); (690) generates patent entities (PE) through the patent leading equation (181) and the first patent indicator data (121P). a patent leading score (191); (700) based on the patent leading score (191) and the first financial indicator data (121F), generating a predicted value of the compounded-price yield of each patent entity (PE) (701); The ranking of the recurrence price yield (701) is sorted as the basis for stock selection.

Another object of the present invention is to provide a patented big data predictive stock selection method (800) implemented by a computer, comprising the following steps: (810) setting parameters: multiple patent entities (PE), one time lead (L), one First year (T1), one second year (T2), multiple patent indicators (PI) and a finance used to describe each patent entity (PE) in the first year (T1) and the second year (T2) Indicators (FI), patent indicators (PI) include indicators describing invention patents, indicators describing invention patents, indicators describing utility model patents, Describe the indicators of the design patent and the indicators describing the valid patents. Among them, the patent entity (PE) is a listed company, and the financial indicator (FI) is the patent entity (PE) in the first year (T1) and the second year (T2). ) The closing price (SP) of the last trading day in the first trading day (T1) and the second year (T2) are both one year, and the end of the second year (T2) (T20) is the same. The end of the year (T1) (T10) is further behind a time lead (L), the time lead (L) is a quarter; (820) Collecting data: each patent entity (PE) in the first year (T1) Within, the first patent indicator (PI) and the financial indicator (FI) correspond to a plurality of first patent indicator data (121P) and a plurality of first financial indicator data (121F), collecting each patent entity (PE) in the second year ( Within T2), a plurality of second patent indicator data (122P) corresponding to the patent indicator (PI) and the financial indicator (FI) and a plurality of second financial indicator data (122F); (830) the first patent indicator data ( 121P), first financial indicator data (121F), second patent indicator data (122P), and second financial indicator data (122F) constitute a first panel data (131); (840) provide A normal distribution conversion program (141) converts the first panel data (131) into a second panel data (142); (850) provides a one-time Granger causality test model based on the time lead (L) (151) At least one first fitting coefficient (152), and a first threshold (153) corresponding to the first fitting coefficient (152), the self-variable of the unary Granger causality test model (151) is the second panel A patent indicator (PI) of the data (142), the variable is the financial indicator (FI) of the second panel data (142); (860) successively uses the one-dimensional Granger causality test model (151) to calculate the second panel data (142), a plurality of patent core indicators (PCI) are selected from a plurality of patent indicators (PI), and a first fitting coefficient (152) of each patent core indicator (PCI) meets a first threshold (153); 870) providing a time lead (L) based multivariate Granger causality test model (171), at least one second fit coefficient (172), and a second threshold corresponding to the second fit coefficient (172) ( 173), the independent variable of the multivariate Granger causality test model (171) is all patent core indicators (PCI) in the second panel data (142), and the variable is the second Data plate (142) financial indicators (FI); (880) by multivariate Granger causality test model (171), generates a calculation Patent collar The first equation (181), the patent leading equation (181) consists of multiple patent leading indicators (PLI) and the corresponding weighting factors (182) of the patent leading indicators (PLI). The patent leading indicator (PLI) consists of the patent core indicators ( PCI) selected, in the patent leading equation (181), the second fitting factor (172) of each patent leading indicator (PLI) met the second threshold (173); (890) passed the patent leading equation (181) and The first patent indicator data (121P) generates a patent lead score (191) for each patent entity (PE), and the patent lead score (191) is the predicted return rate (SPR) predicted value of each patent entity (PE); And (900) sort the patent lead score (191) as a stock selection basis.

It is still another object of the present invention to provide a computer system (400) for patent big data forecasting stock selection for implementing the aforementioned patent big data forecasting stock selection method (100, 600, 800) realized by a computer. The patented big data predictive stock selection computer system (400) includes: an indicator calculation unit (420) for calculating the patent index (PI) and financial indicator (FI) data of the patent entity (PE) to generate the first patent indicator Data (121P), first financial indicator data (121F), second patent indicator data (122P) and second financial indicator data (122F); a database unit (430) for storing information of the patent entity (PE) First patent indicator data (121P), first financial indicator data (121F), second patent indicator data (122P) and second financial indicator data (122F); a calculation and prediction unit (440) for calculating patents Core Indicators (PCI), Patent Leading Indicators (PLI), Patent Leadership Formula (181), and Patent Leadership Scores (191); a Display and Export Unit (450) for Presenting Patent Entities (PEs) and Their Corresponding Patent Leaders A score (191); and a core control unit (410) for integrating and manipulating the above units (420, 430, 440, 450).

The patent big data forecasting stock selection method and computer system implemented by the invention are objective quantitative models, objective and rigorous, and through computer calculation, no human intervention. Not only is it particularly suitable for various patent materials in mainland China, including invention patents, invention patents, and real patents. It uses new patents, design patents, and valid patents. It is also suitable for patents of other countries, and mines patent core indicators, patent leading indicators, and patent leading equations with leading financial information. For the stock selection investment of technology stocks, the evaluation time can be greatly reduced, the stock picking income can be effectively improved, and the positive development of the investment method in the investment field can be promoted, and the research and development and innovation of industrial technology can be positively supported.

Other specific technical features and non-obvious outstanding effects relating to the present invention will be described in detail in the following sections.

100, 600, 800‧‧‧ Patented big data forecasting stock selection method realized by computer

110, 120, 130, 135, 140, 150, 160, 170, 180, 190, 200, 610, 620, 630, 640, 650, 660, 670, 680, 690, 700, 710, 810, 820, 830, 840, 850, 860, 870, 880, 890, 900 ‧ ‧ steps

PE‧‧‧ patent entity

L‧‧‧ time lead

T1‧‧‧ first time period

T2‧‧‧ second time period

PI‧‧‧ patent indicators

FI‧‧‧ financial indicators

121P‧‧‧First Patent Indicator Data

121F‧‧‧First financial indicator data

122P‧‧‧Second patent indicator data

122F‧‧‧Second financial indicator data

131‧‧‧First panel data

141‧‧‧Normal distribution conversion procedure

142‧‧‧ second panel data

151‧‧‧First time series calculation program

152‧‧‧First fitted coefficient

153‧‧‧ first threshold

PCI‧‧‧ patent core indicators

171‧‧‧Second time series calculation program

172‧‧‧second fit factor

173‧‧‧ second threshold

181‧‧‧ Patent Leadership Formula

191‧‧‧ Patent Leading Score

T20‧‧‧ End of the second time period

T10‧‧‧ End of the first time period

SP‧‧‧Resumption closing price

701‧‧‧Reward price yield forecast

SPR‧‧‧Recovery closing price yield

400‧‧‧Public big data forecasting computer system for stock selection

420‧‧‧ indicator calculation unit

430‧‧‧Database unit

440‧‧‧calculation and prediction unit

450‧‧‧Display and export unit

410‧‧‧Core Control Unit

1 is a first preferred embodiment of the present invention, which is a flowchart of a patent big data prediction stock selection method (100) implemented by a computer; FIG. 2 is a first time period (T1) and a second time period ( Schematic diagram of T2) and time lead period (L); FIG. 3 is a second preferred embodiment of the present invention, which is a flowchart of a patent big data prediction stock selection method (600) realized by a computer; The third preferred embodiment of the present invention is a flowchart of a patent big data prediction stock selection method (800) implemented by a computer; FIG. 5 is a fourth preferred embodiment of the present invention, which is a patent big data prediction. The computer system for selecting stocks (400); FIG. 6 is a schematic diagram of a second panel data table proposed by the present invention; FIG. 7 is a schematic diagram of a patent core index proposed by the present invention; FIG. 8 is a patent leading index and corresponding weights proposed by the present invention. Schematic diagram of the coefficient; FIG. 9 is a schematic diagram of the patent lead score proposed by the present invention; FIG. 10 is a schematic diagram of the stock selection with the patent lead score.

The invention mainly discloses a patent big data forecasting stock selection method and application realized by a computer, wherein the basic knowledge of patent materials, patent indexes, financial indicators, etc., has been understood by those having ordinary knowledge in the related technical field, The description below is not fully described. In the meantime, the drawings referred to hereinafter are merely illustrative of the features of the present invention, and are not necessarily required to be completely drawn according to actual dimensions, as previously described.

Referring to FIG. 1, a first preferred embodiment of the present invention is a patent big data prediction stock selection method (100) implemented by a computer, which includes the following steps:

Step 110: Set parameters, including: multiple patent entities (PE), a time length (T0), a time lead period (L), a first time period (T1), and a second time period (T2), To describe a plurality of patent indicators (PIs) and at least one financial indicator (FI) of each patent entity (PE) over a length of time (T0), wherein the first time period (T1) and the second time period (T2) have The same length of time (T0), the end of the second time period (T2) (T20) is one time behind the end of the first time period (T1) (T10) (L).

Step 120: Collect data, including: a plurality of first patent indicator data (121P) corresponding to patent indicators (PI) and financial indicators (FI) of each patent entity (PE) in a first time period (T1) First financial indicator data (121F), collecting multiple second patent indicator data (122P) corresponding to patent indicators (PI) and financial indicators (FI) of each patent entity (PE) in the second time period (T2) ) with multiple second financial indicator data (122F).

Step 130: The first patent indicator data (121P), the first financial indicator data (121F), the second patent indicator data (122P), and the second financial indicator data (122F) are combined into a first panel data (131).

Step 140, providing a normal distribution conversion program (141), and forming a second panel data (142) by passing the first panel data (131) through the normal distribution conversion program (141).

Step 150, providing a first time series operation program (151) based on the time lead (L), at least one first fitting coefficient (152), and a first threshold corresponding to the first fitting coefficient (152) (153), the self-variable of the first time series operation program (151) is a patent indicator (PI) of the second panel data (142), and the variable is the financial indicator (FI) of the second panel data (142).

In step 160, the first time series operation program (151) is used successively, and the second panel data (142) is calculated, and the first fitting coefficient (152) is selected from the plurality of patent indicators (PI) to meet the first threshold (153). ) Multiple patent core indicators (PCI).

Step 170, providing a second time series operation program (171) based on the time lead (L), at least one second fitting coefficient (172), and a second threshold corresponding to the second fitting coefficient (172) (173), the independent variable of the second time series operation program (171) is all the special data in the second panel data (142) The core indicator (PCI), because the variable is the financial indicator (FI) of the second panel data (142).

Step 180, through the second time series operation program (171), the calculation generates a patent leading equation (181), and the patent leading equation (181) is weighted by a plurality of patent leading indicators (PLI) and patent leading indicators (PLI). The coefficient (182) is composed, and the patent leading indicator (PLI) is selected by the patent core index (PCI). In the patent leading equation (181), the second fitting coefficient (172) of each patent leading indicator (PLI) is consistent. The second threshold (173).

In step 190, the first patent indicator data (121P) is introduced into the patent leading equation (181) to generate a patent lead score (191) of each patent entity (PE).

In step 200, the stock selection is based on the patent lead score (191).

In the above step 110, the patent entity (PE) is a patent entity that has the patent right and can profit from the operation of the patent right, preferably a publicly listed listed company, such as Shanghai Main Board Company, Shenzhen Main Board Company, Shenzhen Small and Medium Board Company, Shenzhen The GEM company, but not limited to the listed company; this embodiment is also applicable to non-listed companies, such as the company within the National Small and Medium Enterprise Share Transfer System (referred to as the New Third Board), as long as it can accept external funds to enter, share equity and The subject of the equity proceeds is the scope of application of this embodiment.

Moreover, the patents referred to in step 110 are not limited to authorized patents, as long as they are patents published in the patent database, including invention patents, invention patents, utility model patents, design patents, and patent rights. Patent, patent invalid patents, etc. At the same time, the method proposed in this embodiment can effectively solve the problem that the information content of the Chinese mainland patent is different from the US patent, and can be applied to patents of various regions in the world.

In this embodiment, the patent index (PI) is preferably a patent quantity indicator that can be automatically calculated by a computer in an annual calculation unit, including patent indexes of all patents within one year before the end date of the time period, and two years before the end date of the time period. The patent indicators of all patents, the patent indicators of all patents within 3 years before the end of the time period, and the like; and the patent indicators of all valid patents within 1 year before the end of the time period, all within 2 years before the end of the time period The patent index of a valid patent, the patent index of all valid patents within 3 years before the end of the time period, and the like. The end of the time period includes the first time period (T1) The end date (T10) and the end date of the second time period (T2) (T20).

Take the patents in mainland China as an example. The patent indicators of all patents within one year before the end of the time period include:

P101: Number of invention patents published one year before the end date of the current period

P102: Number of utility model patents published one year before the end date of the current period

P103: Number of design patents published one year before the end date of the current period

P104: Number of invention patents published one year before the end date of the current period

P105: The average number of invention patents granted by the invention one year before the end date of the current period, the definition of the examination period: patent application date to patent authorization date, unit: year

P106: Total number of invention patents IPC classification numbers 1 year before the end date of the current period

P107: Total number of utility model patent IPC classification numbers 1 year before the end date of the current period

P108: Total number of IPC classification numbers for invention patents issued one year before the end date of the current period

P109: Average number of invention patents IPC classification number 1 year before the end date of the current period

P110: Average number of utility model patent IPC classification numbers 1 year before the end date of the current period

P111: Average number of IPC classification numbers of invention patents issued one year before the end date of the current period

P112: Total number of invented patent specifications 1 year before the end date of the current period

P113: Total number of words in the utility model patent specification for one year before the end date of the current period

P114: Total number of patents for invention patents issued one year before the end date of the current period

P115: The average number of words in the invention patent specification 1 year before the end date of the current period

P116: Average number of words in the utility model patent specification 1 year before the end date of the current period

P117: Average number of words for invention patents issued one year before the end date of the current period

P118: Total number of claims for invention patents 1 year before the end date of the current period

P119: Total number of claims for utility model patents 1 year before the end date of the current period

P120: Total number of claims for invention patents for one year before the end date of the current period

P121: Average number of claims for invention patents 1 year before the end date of the current period

P122: Average number of claims for utility model patents 1 year before the end date of the current period

P123: Average number of claims for invention patents 1 year before the end date of the current period

P124: Total number of exclusive patents for invention patents 1 year before the end date of the current period

P125: Total number of exclusive patents for utility model patents one year before the end date of the current period

P126: The total number of exclusive patents for invention patents 1 year before the end date of the current period

P127: Average number of unique patents for invention of patents one year before the end date of the current period

P128: Average number of exclusive items of utility model patents for one year before the end date of the current period

P129: The average number of unique claims for invention patents 1 year before the end date of the current period

P130: Total number of drawings for inventing the patent specification 1 year before the end date of the current period

P131: Total number of drawings of utility model patent specification for one year before the end date of the current period

P132: Total number of drawings for invention patents issued one year before the end date of the current period

P133: Average number of drawings of the invention patent specification 1 year before the end date of the current period

P134: Average number of drawings of utility model patent specification for one year before the end date of the current period

P135: Average number of drawings for invention patents issued one year before the end date of the current period

P136: Total number of inventions in the first year before the end of the current period

P137: Total number of words in the utility model summary 1 year before the end of the current period

P138: Total number of invention authorization summaries 1 year before the end date of the current period

P139: Average number of inventions in the first year before the end of the current period

P140: Average number of utility model abstracts 1 year before the end of the current period

P141: Average number of invention authorization abstract words 1 year before the end date of the current period

P145: Total number of patents 1 year before the end date of the current period = P101 + P102 + P103 + P104

P146: Proportion of invention patents 1 year before the end date of the current period = P101/P105

P147: Proportion of utility model patents 1 year before the end date of the current period = P102/P105

P148: Proportion of design patents 1 year before the end date of the current period = P103/P105

P149: Proportion of invention patents for one year before the end date of the current period = P104/P105

P150: Average number of life claims of inventions in the first year before the end date of the current period

P151: Average number of utility model patents for one year before the end date of the current period

P152: Average number of life expectancy of design patents 1 year before the end date of the current period

P153: Average number of life claims for invention patents 1 year before the end date of the current period

Taking valid patents as an example, the patent indicators of all valid patents within one year before the end of the time period include:

PA101: Number of patents valid for inventions published one year before the end date of the current period

PA102: Number of valid patents for utility models published one year before the end date of the current period

PA103: Number of patents valid for design published one year before the end date of the current period

PA104: Number of valid patents granted by the invention one year before the end date of the current period

PA105: Average number of valid patent examination periods for inventions granted one year before the end date of the current period

PA106: The total number of valid patents IPC classification numbers for inventions one year before the end date of the current period

PA107: Total number of valid patent IPC classification numbers for utility models 1 year before the end date of the current period

PA108: The total number of valid patent IPC classification numbers for invention authorization 1 year before the end date of the current period

PA109: The average number of patents validated by the IPC classification number of inventions in the first year before the end date of the current period

PA110: Average number of valid patent IPC classification numbers for utility models 1 year before the end date of the current period

PA111: The average number of valid patent IPC classification numbers for invention licenses 1 year before the end date of the current period

PA112: The total number of patents for valid invention patents issued one year before the end date of the current period

PA113: The total number of valid patent specifications for utility models for one year before the end date of the current period

PA114: The total number of words of the invention patent authorization valid patent specification one year before the end date of the current period

PA115: The average number of words of the invention patent valid for one year before the end date of the current period

PA116: Average number of valid patent specifications for utility models for one year before the end of the current period

PA117: The average number of words of the invention patent authorization valid patent specification one year before the end date of the current period

PA118: Total number of claims for invention of valid patents one year before the end date of the current period

PA119: Total number of claims for valid patents for utility models 1 year before the end date of the current period

PA120: The total number of claims for the invention of a valid patent for one year before the end date of the current period

PA121: Average number of claims for invention of valid patents one year before the end date of the current period

PA122: Average number of claims for valid patents for utility models 1 year before the end date of the current period

PA123: The average number of claims for the invention of a valid patent for one year before the end of the current period

PA124: The total number of exclusive patents for the invention of valid patents one year before the end date of the current period

PA125: The total number of exclusive patents for valid patents for utility models 1 year before the end date of the current period

PA126: The total number of exclusive patents for the invention of a valid patent for one year before the end of the current period

PA127: The average number of independent patents for inventing valid patents one year before the end date of the current period

PA128: The average number of unique items of valid patents for utility models 1 year before the end of the current period

PA129: The average number of unique items of invention patents granted for one year before the end date of the current period

PA130: Total number of drawings for inventing valid patent specifications 1 year before the end date of the current period

PA131: Total number of drawings of valid patent specifications for utility models for one year before the end date of the current period

PA132: Total number of drawings for the invention of a valid patent specification for one year before the end date of the current period

PA133: Average number of drawings for inventing valid patent specifications one year before the end date of the current period

PA134: Average number of drawings of valid patent specifications for utility models for one year before the end of the current period

PA135: Average number of drawings for the invention of a valid patent specification for one year before the end date of the current period

PA136: The total number of words invented in the first year before the end of the current period

PA137: The total number of words in the utility model summary 1 year before the end of the current period

PA138: Total number of invention authorization summaries 1 year before the end date of the current period

PA139: Average number of inventions in the first year before the end of the current period

PA140: The average number of words in the utility model summary 1 year before the end of the current period

PA141: Average number of invention authorization abstract words one year before the end date of the current period

PA145: The total number of valid patents for one year before the end date of the current period

PA146: Proportion of invention effective patents one year before the end date of the current period

PA147: Proportion of valid patents for utility models 1 year before the end of the current period

PA148: Proportion of effective patents for design one year before the end date of the current period

PA149: Proportion of valid patents for inventions for one year before the end date of the current period

PA150: The average number of valid patent lifespans for inventions one year before the end date of the current period

PA151: Average number of valid patent life for utility models 1 year before the end date of the current period

PA152: Average number of valid patent life for design 1 year before the end date of the current period

PA153: Average number of valid patent lifetimes for inventions for one year before the end date of the current period

PA154: Average number of valid patent examination periods for invention authorization 1 year before the end date of the current period

PA155: Proportion of invention disclosure in valid patent = PA101/PA145

PA156: Proportion of utility model in valid patent = PA102/PA145

PA157: Proportion of design in valid patents = PA103/PA145

PA158: Proof of invention in valid patents = PA104/PA145

As for the patent indicators of all patents within 2 years, the first 3 years, the first 4 years, the first 5 years, and the like before the end of the time period; and within 2 years, the first 3 years, and the first 4 years before the end of the time period The patent indicators of all valid patents within the first 5 years, and so on, as long as the length of time is extended, the calculation formula is the same. Preferably, the embodiment adopts a total of 1040 patent indicators (PI) within one year before the end of the time period to the first 10 years, including 540 valid patent indicators.

Not every patent indicator (PI) can effectively predict financial indicators (FI). Only the patent core indicator (PCI) and patent leading indicator (PLI) can effectively predict financial indicators (FI), which requires rigorous statistical analysis. Verification and error verification, which is the core of this embodiment, will continue to be explained in the subsequent pages.

In terms of financial indicators (FI), the examples used in this example are indicators that express the business performance of the enterprise. They can be solvency indicators, operational capability indicators, return on the common stockholders (ROE), assets. Rate of Return on Assets (ROA), EPS (Earnings Per Share), Price-to-Book Ratio (MTB), stock price-related indicators, etc., which are not limited in this embodiment, but are preferred. It is a stock price related indicator, such as: closing price, reclosing closing price (SP), and reclosing closing price (SPR).

In the above step 110, the units of the first time period (T1) and the second time period (T2) may be monthly, quarterly, semi-annual, or annual, etc., but preferably one year. Because in practice, the number of patent disclosures in a year is not uniform every month, but there are seasonal changes, some have more months, and some have fewer months. If you take quarterly or monthly units, you need to spend more calculations to eliminate seasonal changes. Factors; using the annual, can minimize the impact of seasonal changes.

The first time period (T1) and the second time period (T2) are set to collect sufficient sample data for model and verification purposes. Since the present embodiment proposes a prediction model, that is, the data of the previous period is used to predict the current period data, or the data of the current period is predicted by the current period data, so at least two periods of data are required to establish the model and Verify the significance of the predicted results.

Please refer to FIG. 2. In this embodiment, the first time period (T1) and the second time period (T2) have the same length of time (T0), and the end date (T10) of the first time period (T1) is the deadline. (D), the end of the second time period (T2) (T20) is one time behind the end of the first time period (T1) (L). The time lead period (L) may be one month, one quarter, half year, or one year; etc.; considering the frequency of publication of the patent data and the investment habit of the investment institution, the time lead period (L) of the present embodiment is preferably one quarter. As can be seen from Figure 2, the time span of the first time period (T1) and the second time period (T2) has a partial overlap. For example, if the lead time (L) is one quarter, the first time period (T1) is from October 1, 2014 to September 30, 2015, and the end of the first time period (T1) (T10), The second time period (T2) is from July 1, 2014 to June 30, 2015, and the end date (T20) of the second time period (T2) is June 30, 2015.

In the above step 120, according to the set time lead period (L), the first time period (T1) and the second time period (T2), the computer entity starts collecting and calculating the patent entity (PE) in the first time period ( The first patent indicator data (121P) and the first financial indicator data (121F) in T1), and the second patent indicator data (122P) and second finance in the second time period (T2) of the patent entity (PE) Indicator data (122F).

In step 130, the panel data in the first panel data (131), also referred to as parallel data or comprehensive data, is a mixture of time series data and cross-sectional data, and refers to the number of M cross-section objects being observed in the time period. (N) The data set has a total of M × N data sets. Taking this embodiment as an example, there is a first time period (T1) and a second time period (T2), that is, N=2, and there are two data collection periods; if there are 1000 patent entities (PE), that is, M=1000 At this time, 1000 data sets of cross-section observation objects, financial indicators (FI) and 1040 patent indicators (PI) in two time periods are formed.

Traditional time series data is used to analyze a single observed object at multiple times The observations (self-variables and dependent variables) are related. Traditional cross-sectional data is used to analyze the observations of multiple observed objects at a single point in time (self-variables and dependent variables). Both of these data are not applicable to the method proposed in this embodiment, because the embodiment has a plurality of observed objects and has a plurality of time points, and each time point has a plurality of independent variables and dependent variables. The panel data is used to analyze the correlation between observations of multiple specific cross-section objects at multiple time points. Due to the increase of observation values, the sampling accuracy of the estimators can be increased, and more consistent estimators can be obtained. With the effective estimation amount and obtaining more dynamic information, the present embodiment uses panel data for analysis.

Between the above steps 130 and 140, a normal distribution test program (135) may be further (but not necessarily), and each patent indicator data and each financial indicator data in the first panel data (131) may be included. , check the state of its normal distribution. Because the data does not exhibit a normal distribution, the direct impact is modeling difficulties. In the process of modeling calculus, the model is often collapsed due to the high error and the inability to pass the statistical verification requirements. Therefore, prior to modeling, the normal distribution test program (135) is pre-executed on the first panel data (131) to detect its data distribution state, which can facilitate the normal distribution conversion process (141) of the subsequent step 140.

The distribution curve of a normal distribution is basically a curve in which a center line is bilaterally symmetrical at an expected value (average value) and extended to both sides in units of standard deviation. The normal distribution test program (135) is commonly used in the following ways: Anderson-Darling test program, Ryan-Joiner test program, Kolmogorov-Smirnov test program, etc., or it is easier to observe Skewness and kurtosis. The coefficient (Kurtosis) can infer the normal distribution of the data, and this embodiment does not limit which test program is used.

In the above step 140, the Box-Cox conversion program is a commonly used normal distribution conversion program (141), but the embodiment is not limited thereto. The first panel data (131) of the present embodiment is converted into the normally distributed second panel data (142) by the normal distribution conversion program (141).

In step 150, the first time series operation program (151) is preferably a one-dimensional Granger Causality Test Model. The Granger causality test model was created by Clive W.J. Granger, winner of the 2003 Nobel Prize in Economics, for the analysis of time series. Leading and backward relationship between economic variables. The basic idea is that if there are two variables X and Y, the variable X occurs first, the variable Y occurs after, and after the Granger causality test model, the probability that the variable X is valid for the variable Y is significant. The effect is called the variable X is ahead of the variable Y, or the variable X is the leading indicator of the variable Y. In the economic variables processed by the Granger causality test model, the independent variables and the dependent variables are time series variables, and the basic operation model is the regression analysis model, but before the regression analysis, the self-variables and the dependent variables are set for a period of time. The offset, that is, the set time lead (L) or time lag period, and then the regression analysis model with lead or backward period to check the matching effect to verify the leading effect or backward effect of the independent variable.

After setting the appropriate time lead (L), measure the prediction effect of the Granger causality test model, and must consider the model fit, that is, the first fitting coefficient (152) of this embodiment and its corresponding first threshold ( 153). The first fitting coefficient (152) can be obtained by the F-test, and the p-value is generally <0.1, which is acceptable for the model; if the p-value is <0.05, the model is good; if the p-value is <0.005, it is excellent. Model; at this time, the first threshold (153) can be set to be no more than 0.1. Simply put, if the self-variable is X, the variable is Y, and the p-value <0.05 after the Granger causality test model indicates that the self-variable X is dependent on the variable in the 95% confidence interval. For Y, has a leading effect. The first fitting coefficient (152) in this embodiment may also use the absolute value of the t-test value. The larger the absolute value of the t-test value, the better. In the normal ideal case, the absolute value of the t-test value should be no less than 2, so The first threshold (153) may be set to be not less than 2.

In step 150, the setting of the first threshold (153) is critical, the first threshold (153) is set too strict, and in step 160, the patent core indicator (PCI) that is significant cannot be unearthed; A threshold (153) is set too loosely, and a patent indicator (PI) that may be too insignificant may also be dug up in step 160 by being mistaken for the patent core indicator (PCI). If the first time series operation program (151) adopts a one-dimensional Granger causality test model, as described above, if the first fitting coefficient (152) uses the p-value obtained by the F-test, we can set the first The threshold (153) is 0.1. First, we can first understand how many patent core indicators (PCI) can reach the 90% confidence interval. If the number of patent core indicators (PCI) is small, set the first threshold (153) to 0.1; If the number of patent core indicators (PCI) is large, you can set the first threshold (153) to 0.05 or 0.005 to dig out more critical and leading effects more significantly. Patent Core Indicators (PCI).

In the Granger causality test model, the number of independent variables is not limited. That is, the Granger causality test model can analyze the leading effect of an independent variable on a dependent variable, which is called the one-dimensional Granger causality test model, as in step 160; it can also analyze multiple independent variables to one dependent variable simultaneously. The leading effect, at this time called the multivariate Granger causality test model. But we must understand that if there is severe collinearity between multiple independent variables, or the data dispersion of the independent variables is too high, the multivariate Granger causality test model is often easy to collapse in the operation process. Therefore, in step 160, preferably, the patent indicator data (121P, 122P) of the individual patent indicator (PI) is first applied to the financial indicator data (121F, 122F) of the individual financial indicator (FI). The one-dimensional Granger causality test model operation, which extracts the individual patent indicators (PI) with significant leading effects from 1040 patent indicators (PI), called the patent core index (PCI), and excludes other The leading effect does not have a significant patent indicator (PI). At this time, each of the patent core indicators (PCI) excavated has a significant effect on the financial indicators (FI). By significance, it is meant that the first fit factor (152) conforms to the first threshold (153).

In step 170, the present embodiment analyzes multiple patent core indicators (PCI) and financial indicators using a second time series operation program (171) and a time lead period (L), preferably a multivariate Granger causality test model. (FI) leading effect.

In step 180, the second time series operation program (171) of the embodiment is a multivariate Granger causality test model, and the self-variable is a plurality of patent core indicators (PCI) generated in step 160, and the variable is a financial indicator. (FI). Through step 180, a patent leading indicator (PLI) can be extracted from a plurality of patent core indicators (PCI), and a patent leading indicator (PLI) algorithm can be generated to generate a patent leading equation (181). The patent leading equation (181) contains a number of patent leading indicators (PLI) and a linear combination of each patent leading indicator (PLI) with its corresponding weighting factor, as shown below, which is a patent consisting of n patent leading indicators. Leading equation (181): w1 x PLI_1+w2 x PLI_2+w3 x PLI_3+...........+wn x PLI_n

Among them, PLI_1, PLI_2, PLI_3, and PLI_n are patent leading indicators (PLI); w1, w2, w3, and wn are the corresponding weight coefficients of each patent leading indicator (PLI).

Through the patented leading equation (181), it will be faster and more convenient than simply using the patent core indicator (PCI) or the patent leading indicator (PLI) to predict the financial performance of the company. Because of the patent core indicator (PCI) obtained in step 160, each patent core indicator (PCI) is significant for leading financial and financial indicators (FI). Which patent core indicator (PCI) is better? It is a problem. If you want to combine these patent core indicators (PCI), how to set the weighting factor of each patent core indicator (PCI) when combining? Step 180 is to solve these problems. Because the multivariate Granger causality test model, the time lead (L), and the second fit coefficient (172) meet the second threshold (173), a patent leading indicator (PLI) can be generated corresponding to each other. Weight coefficient (182).

The second fitting coefficient (172) may be the same as the first fitting coefficient (152) described above, and the p-value of the F-test may be used, or the t-test value may also be used. If the first fitting coefficient (152) uses the p-value of the F-test, the second threshold (173) may be set to be no more than 0.1; if the first fitting coefficient (152) uses the absolute value of the t-test value, The second threshold (173) may be set to be not less than 2.

In step 180, we must understand another important concept. The multivariate Granger causality test model is not a simple summation of the results of multiple unary Granger causality test models, when multiple patent core indicators (PCI) are combined. At the same time, the significance of the leading effect of the individual patent core indicators (PCI) on financial indicators (FI) will change, and even the leading effect of some patent core indicators (PCI) will become less significant. Therefore, in step 180, preferably, the self-variable item-by-item deletion program can be further operated. That is to say, all the patent core indicators (PCI) are first included in the multivariate Granger causality test model, and the second fitting coefficient (172) and the second threshold (173) are set to observe the core indexes of each patent after the calculation. (PCI)'s second fitting factor (172), removing the least significant or even insignificant patent core indicator (PCI), and then redoing the remaining patent core indicator (PCI) into a multiple Granger causality test model Then delete the least significant patent core indicator (PCI), repeat this process, and finally leave the second fitting coefficient (172) in line with the second threshold (173), with a significant patent core indicator (PCI). Called the Patent Leading Indicator (PLI). At this point, the multivariate Granger causality test model integrates all patent leading indicators (PLI) to generate a patented leading equation (181). The patent leading equation (181) consists essentially of multiple patent leading indicators (PLI) and their corresponding weights. The coefficient (182) is composed.

In step 190, the first patent indicator data (121P) of the patent entity (PE) is input into the patent leading equation (181), which can generate a patent leading score (191) of each patent entity (PE), and the patent leading score ( 191) is a predicted value representing the financial indicator (FI) predicted value of the patent entity (PE) after the next time lead (L) of the end of the first time period (T1).

In step 200, the higher the patent lead score (191), the financial indicator corresponding to the patent entity (PE) after the time lead (L) from the end of the first time period (T1) (T10) ( FI) The higher the predicted value; the lower the patent lead score (191), the lower the financial indicator (FI) predicted on behalf of the patent entity (PE). Since the forecast value of the company's financial indicators (FI) directly expresses its future business performance, the better the company's future business performance, the more investment value. Therefore, observing the high or low of the patent leading score (191), it is possible to select objects with investment potential from the patent entity (PE).

Please refer to FIG. 4 , which is a second preferred embodiment of the present invention. The method for predicting stock selection of a patent big data implemented by a computer ( 600 ) includes the following steps:

Step 610: Set parameters, including: a plurality of patent entities (PE), a time lead period (L), a first year (T1), and a second year (T2), which are used to describe each patent entity (PE). Multiple patent indicators (PI) and a financial indicator (FI) in the first year (T1) and the second year (T2), wherein the patent entity (PE) is a listed company, the first year (T1) and the second The length of the year (T2) is one year, and the end of the second year (T2) (T20) is one time behind the end of the first year (T1) (T10). The indicator (FI) is the reinstatement closing price (SP) of each patent entity (PE) on the last trading day in the first year (T1) and the second year (T2), and the time lead period (L) is one quarter, the patent Indicators (PI) include indicators describing invention patents, indicators describing invention patents, indicators describing utility model patents, indicators describing design patents, and indicators describing effective patents.

The so-called reinstatement is to repair the stock price and volume, and adjust the trading volume to the same equity. For example, a stock is originally 20 per share, 10 shares are sent to 10 shares, and after deducting the rights, it becomes 10 shares per share. The stock price is halved, but the share capital is doubled. At this time, the reinstatement price is 20 yuan per share. If the stock price now becomes 13 blocks, the re-emphasis price is 26 per share. By reinstating, it is possible to eliminate the distortion of the stock price caused by the ex-dividend ex-dividend and to observe the performance of the patent entity (PE).

Step 620: Collect data, including: a plurality of first patent indicator data (121P) and a plurality of patent indicators (PI) and financial indicators (FI) in each patent entity (PE) in the first year (T1) The first financial indicator data (121F) collects multiple second patent indicator data (122P) corresponding to each patent entity (PE) in the second year (T2), patent indicator (PI) and financial indicator (FI). Multiple second financial indicator data (122F).

Step 630, the first patent indicator data (121P), the first financial indicator data (121F), the second patent indicator data (122P), and the second financial indicator data (122F) are combined into a first panel data (131).

Step 640, providing a normal distribution conversion program (141) to convert the first panel data (131) into a second panel data (142).

Step 650, providing a one-time Granger causality test model (151) based on a time lead (L), at least one first fitting coefficient (152), and a first threshold corresponding to the first fitting coefficient (152) (153), the independent variable of the unary Granger causality test model (151) is a patent indicator (PI) of the second panel data (142), and the variable is the financial indicator (FI) of the second panel data (142).

In step 660, the first-order Granger causality test model (151) is used successively, the second panel data (142) is calculated, and a plurality of patent core indicators (PCI) are selected from a plurality of patent indicators (PI), and the core indexes of each patent are obtained. The first fit factor (152) of (PCI) conforms to the first threshold (153).

Step 670, providing a time lead (L) based multivariate Granger causality test model (171), at least one second fit coefficient (172), and a second threshold corresponding to the second fit coefficient (172) (173), the independent variable of the multivariate Granger causality test model (171) is all patent core indicators (PCI) in the second panel data (142), and the variable is the financial indicator of the second panel data (142) (FI) ).

Step 680, through the multivariate Granger causality test model (171), the calculus generates a patent leading equation (181), and the patent leading equation (181) is determined by a plurality of patent leading indicators (PLI) and patent leading indicators (PLI). The weighting factor (182) is composed, and the patent leading indicator (PLI) is selected by the patent core index (PCI). In the patent leading equation (181), the second fitting coefficient (172) of each patent leading indicator (PLI) is A second threshold (173) is met.

In step 690, a patent lead score (191) of each patent entity (PE) is generated by the patent leading equation (181) and the first patent indicator data (121P). This patent lead score (191) is a predictive value representing the predicted return price (SP) of the patent entity (PE) after the next lead (L) of the first year (T1) end date (T10).

Step 700, based on the patent lead score (191) and the first financial indicator data (121F), generate a predicted value of the compound price return rate (701) of each patent entity (PE).

In step 710, the predicted value of the return-to-weight ratio (701) is sorted as a stock selection basis.

The difference between the second preferred embodiment and the first preferred embodiment lies mainly in further defining, as follows:

(1) The limited patent entity (PE) is a listed company.

(2) The length of the first year (T1) and the second year (T2) is limited to one year, which facilitates the collection of the first patent indicator data (121P) and the second patent indicator data (122P).

(3) The limited time lead period (L) is one quarter, which is more in line with investment forecasting habits;

(4) The defined financial indicator (FI) is the reinstatement closing price (SP) of the patent entity (PE) on the last trading day of the first year (T1) and the second year (T2); the reclosing closing price (SP) ) has a more intuitive effect on investment decisions.

(5) Further calculate the predicted value of the re-pricing rate by the patent leading score (191), and limit the stock selection based on the predicted value of the re-valuation rate (701). In this embodiment, the stock selection is not directly carried out by the patent leading score (191), but the stock picking is performed by the predicted value of the compounded rate of return (701). Because the patent leading score (191) represents the replay closing price (SP) forecast, the patent leading score (191) is higher, representing the predicted replay closing price (SP) is higher; the patent leading score (191) is lower, representing The predicted return price (SP) is lower. The key to the performance of investment stock selection is revenue, which is not directly related to the closing price (SP). For example, the predicted stock prices of two stocks are 11 and 101, respectively, and each rises by 1 yuan. It seems that the latter's forecasted stock price is higher, but the former is from 10 to 11 and the yield is 10%. The latter rose from 100 to 101 with a yield of only 1%. In terms of investment, the investment performance of the former is better, and the investment performance of the latter is poor. Therefore, in this embodiment, it does not directly pass the patent leading score (191). The stock picking is based on the stock price forecast (701).

Please refer to FIG. 5, which is a third preferred embodiment of the present invention. The method for predicting stock selection of a patent big data implemented by a computer (800) includes the following steps:

Step 810, setting parameters, including: a plurality of patent entities (PE), a time lead period (L), a first year (T1), and a second year (T2), which are used to describe each patent entity (PE). Multiple patent indicators (PI) and a financial indicator (FI) in the first year (T1) and the second year (T2), wherein the patent entity (PE) is a listed company, the first year (T1) and the second The length of the year (T2) is one year, and the end of the second year (T2) (T20) is one time behind the end of the first year (T1) (T10). The indicator (FI) is the reinstatement closing rate (SPR) of each patent entity (PE) on the last trading day in the first year (T1) and the second year (T2), and the time lead period (L) is one quarter. The patent index (PI) includes indicators describing the patents of the invention, indicators describing the patents granted by the invention, indicators describing the utility model patents, indicators describing the design patents, and indicators describing the effective patents.

Step 820, collecting data: each patent entity (PE) in the first year (T1), the patent indicator (PI) and the financial indicator (FI) corresponding to the plurality of first patent indicator data (121P) and a plurality of first Financial indicator data (121F), collecting multiple second patent indicator data (122P) and multiple patents (PE) corresponding to patent indicators (PI) and financial indicators (FI) in the second year (T2) Second financial indicator data (122F).

Step 830, the first patent indicator data (121P), the first financial indicator data (121F), the second patent indicator data (122P), and the second financial indicator data (122F) are combined into a first panel data (131).

In step 840, a normal distribution conversion program (141) is provided to convert the first panel data (131) into a second panel data (142).

Step 850, providing a time lead (L) based unary Granger causality test model (151), at least one first fit coefficient (152), and a first threshold corresponding to the first fit coefficient (152) (153), the independent variable of the unary Granger causality test model (151) is a patent indicator (PI) of the second panel data (142), and the variable is the financial indicator (FI) of the second panel data (142).

Step 860, using the one-dimensional Granger causality test model (151) one by one, and computing the second side The board data (142) selects a plurality of patent core indicators (PCI) from a plurality of patent indicators (PI), and the first fitting coefficient (152) of each patent core indicator (PCI) meets the first threshold (153). .

Step 870, providing a time lead (L) based multivariate Granger causality test model (171), at least one second fit coefficient (172), and a second threshold corresponding to the second fit coefficient (172) (173), the independent variable of the multivariate Granger causality test model (171) is all patent core indicators (PCI) in the second panel data (142), and the variable is the financial indicator of the second panel data (142) (FI) ).

Step 880, through the multivariate Granger causality test model (171), the calculus generates a patent leading equation (181), and the patent leading equation (181) is determined by a plurality of patent leading indicators (PLI) and patent leading indicators (PLI). The weighting factor (182) is composed, and the patent leading indicator (PLI) is selected by the patent core index (PCI). In the patent leading equation (181), the second fitting coefficient (172) of each patent leading indicator (PLI) is A second threshold (173) is met.

Step 890, generating a patent lead score (191) of each patent entity (PE) through the patent leading equation (181) and the first patent indicator data (121P), and the patent lead score (191) is the reinstatement of each patent entity (PE). The closing price yield (SPR) forecast.

In step 900, the patent lead score (191) is sorted as a stock selection basis.

The difference between the third preferred embodiment and the second preferred embodiment lies mainly in further definition, as follows:

(1) The defined financial indicator (FI) is the reinstatement closing rate (SPR) of the patent entity (PE) on the last trading day of the first year (T1) and the second year (T2); The price return rate (SPR) can be an annual rate of return, a semi-annual rate of return, or a quarterly rate of return, and does not require a limit, but is preferably an annual rate of return.

(2) Direct stock selection through the patent leading score (191), because the patent leading score (191) represents the predicted value of the returning rate of return (SPR), and the patent leading score (191) is higher, representing the resumption of the forecast. The closing price yield (SPR) is higher; the patent leading score (191) is lower, representing a lower return rate (SPR) of the predicted return. The selection of shares with the patent leading score (191) is more intuitive and convenient for the operation of investment stock selection.

The implementation process of the first preferred embodiment to the third preferred embodiment will be further described below by taking the listed company of the A-share in China as a patent entity (PE).

Listed companies in China's A-share market, including the Shanghai Stock Exchange and the Shenzhen Stock Exchange, can be divided into Shanghai Main Board, Shenzhen Main Board, Small and Medium-sized Board, and Growth Enterprise Market. As of the end of 2015, there were more than 2,820 listed companies, and the number of patent entities (PE) was 2,820.

The present invention is a predictive model for establishing a leading indicator of a patent indicator (PI) for financial indicators (FI), and therefore must consider the subsidiary structure of a listed company. If the financial of the subsidiary is consolidated into the parent company, the patent index (PI) of the subsidiary must also be combined with the parent company. Therefore, 2,820 listed companies must investigate the structure of their subsidiaries.

In terms of financial indicators (FI), we are limited to financial indicators (FI) and are further limited to the reinstatement closing price (SPR) and annual yield. In terms of patent indicators (PI), we use a total of 1040 patent indicators (PI), including 540 valid patent indicators, from one year to the first 10 years before the end of the time period.

The time lead period (L) is set to one quarter, the first year (T1) is from October 1, 2014 to September 30, 2015, and the end date (T10) is September 30, 2015. The second year (T2) is from July 1, 2014 to June 30, 2015, and its end date (T20) is June 30, 2015.

In the selection of valid samples, we must set the financial year (F) for the last trading day of the first year (T1) and the second year (T2), that is, the return rate of return (SPR) is not It is 0, and at least 10 patents must be disclosed in the first year (T1), including invention disclosure, invention authorization, utility model, design and so on.

The 2820 patent entities (PEs) obtained the above criteria and obtained 1,161 final qualifications, that is, 1161 valid samples of patent entities (PE).

1161 patent entities (PEs) will collect 1040 first patent indicator data (121P) and first financial indicator data (121F) for the first year (T1) and 1040 second patents for the second year (T2) After the indicator data (122P) and the second financial indicator data (122F), the first panel data (131) is composed for subsequent analysis.

Next, we use the Box-Cox conversion program as a normal distribution conversion program (141) to convert the first panel data (131) into the second panel data (142). Figure 6 is a partial view of the second panel data (142). In the second panel data (142), the independent variable is 1161 patent entities (PE) in the first year (T1) and the second year (T2) of 1040 patent indicators (PI), due to the variable 1161 patent entities (PE) Financial Indicators (FI) in the first year (T1) and the second year (T2): Recoverable Rate of Return (SPR). Then we use the one-quarter Granger causality test model with the time lead period (L) as the first time series operation program (151), and sequentially test the leading relationship of each independent variable pair dependent variable.

Model Fit We use the p-value of the F-test as the first fit factor (152), and the first threshold (153) we set <0.1, that is, the patent index (PI) vs. the return-to-return yield (SPR) The lead relationship is acceptable and significant, reaching a 90% confidence interval.

After passing the second panel data (142) through the operation of the one-dimensional Granger causality test model, we succeeded in discovering that some patent indicators (PI) have significant significance for the return of the return-to-return yield (SPR). It is the patent core indicator (PCI). In this way, we extracted 124 patent core indicators (PCI) from 1040 patent indicators (PI), some of which are listed in Figure 7.

The above-mentioned patent core indicators (PCI) can be used individually to predict the return-to-return yield (SPR) of a patent entity (PE).

Then we use the multivariate Granger causality test model as the second time series operation program (171), and extract the patent leading index (PCI) from the patent core index (PCI) and form the patent leading equation (181).

In the operation of the multivariate Granger causality test model, the p value is set to the second fitting coefficient (172), the second threshold (173) is the p value <0.1, the operation is independent of the variable deletion program, and finally the patent leading indicator is obtained. (PLI), as shown in Figure 8.

The patented leading equation (181) is also generated by this, and the overall p value reaches 0.00004, which is extremely remarkable. Among them, the patent leading equation (181) = C + Σ weight coefficient x patent leading indicators

We can first verify whether the patented leading equation (181) has a predictive effect. In the first panel data (131), the first financial indicator data (121F) of the first year (T1) is the actual returning rate of return (SPR) of the patent entity (PE), and we find the average value of 0.1171, that is, 11.71%.

At this time, we use the patented leading equation (181) to introduce the second patent indicator data (122P) of the second year (T2) into the patent leading equation (181), and we can get the return rate of the first year (T1). The predicted value of (SPR). According to the ranking of the predicted value of the return-to-return yield (SPR), the top 100 preferred patent entities (PEs) are selected, which is the investment target. Then we find out the actual return rate of return (SPR) of the 100 patent entities (PE) in the first year (T1), and find the average value of 0.2667, which is 26.67%, which is much higher than the total patent entity (PE). The average value is 11.71%, which is more than twice the average value of all patent entities (PE), indicating that the generated patent leading equation (181) is feasible.

The prediction model is established, and then the stock selection is carried out for the future that has not occurred. We operate the patent leading equation (181) and the first patent indicator data (121P) to generate the patent lead score (191) of each patent entity (PE), some of which are shown in Figure 9. The patent lead score (191) is the predicted return rate of return (SPR) of each patent entity (PE).

In Figure 9, the return-to-return yield (SPR) and the patent lead score (191) are both converted by the normal distribution normal distribution conversion program (141). Next, we sorted the patent lead score (191), as shown in Figure 10, as a stock selection basis. The 20 patent entities (PEs) listed in Figure 10, the patent leading index issued by Shenzhen Stock Exchange (stock code 399427), officially updated 20 of the 100 sample stocks on January 3, 2016.

The present invention further provides a fourth preferred embodiment, which is a patented big data predictive stock selection computer system (400) for implementing the first preferred embodiment, the second preferred embodiment and the third preferred embodiment. The patented big data forecasting stock picking method implemented by computer (100, 600, 800).

Please refer to FIG. 5, the patent big data predicting stock selection computer system (400) includes: an indicator calculation unit (420) for calculating the patent index (PI) and financial indicator (FI) data of the patent entity (PE), Generating first patent indicator data (121P), first financial indicator data (121F), second patent indicator data (122P) and second financial indicator data (122F); and a database unit (430) for storing patent entities (PE) information, first patent Indicator data (121P), first financial indicator data (121F), second patent indicator data (122P) and second financial indicator data (122F); a calculation and prediction unit (440) for calculating patent core indicators (PCI) ), Patent Leading Indicator (PLI), Patent Leading Equation (181), and Patent Leading Score (191); a display and export unit (450) for presenting patent entities (PE) and their corresponding patent leading scores (191) And a core control unit (410) for integrating and manipulating the above units (420, 430, 440, 450).

The patented big data forecasting stock selection method (100, 600, 800) and the patent big data forecasting stock selection computer system (400) proposed by the invention are based on big data, objective calculation, and rigorous verification. It not only contributes to the development of technical strength of patent data analysis and utilization, but also promotes the positive development of investment methods in the investment field, and plays a positive supporting effect on the research and development and innovation of industrial technology.

The above description should be understood and implemented by those skilled in the relevant art. The above is only the preferred embodiment of the present invention and is not intended to limit the scope of the invention. Any equivalent changes or modifications made based on the disclosure of the present invention should be included in the scope of the patent application.

100‧‧‧Public big data forecasting stock selection method realized by computer

110, 120, 130, 135, 140, 150, 160, 170, 180, 190, 200 ‧ ‧ steps

PE‧‧‧ patent entity

L‧‧‧ time lead

T1‧‧‧ first time period

T2‧‧‧ second time period

PI‧‧‧ patent indicators

FI‧‧‧ financial indicators

121P‧‧‧First Patent Indicator Data

121F‧‧‧First financial indicator data

122P‧‧‧Second patent indicator data

122F‧‧‧Second financial indicator data

131‧‧‧First panel data

141‧‧‧Normal distribution conversion procedure

142‧‧‧ second panel data

151‧‧‧First time series calculation program

152‧‧‧First fitted coefficient

153‧‧‧ first threshold

PCI‧‧‧ patent core indicators

171‧‧‧Second time series calculation program

172‧‧‧second fit factor

173‧‧‧ second threshold

181‧‧‧ Patent Leadership Formula

191‧‧‧ Patent Leading Score

Claims (10)

A patented big data predictive stock selection method (100) implemented by a computer, comprising the following steps: (110) setting parameters: multiple patent entities (PE), one time length (T0), one time lead period (L), one a first time period (T1), a second time period (T2), a plurality of patent indicators (PI) for describing each patent entity (PE) within the length of time (T0) and at least one financial indicator (FI) ), wherein the first time period (T1) has the same length of time (T0) as the second time period (T2), and the end date (T20) of the second time period (T2) is earlier than the first time The end date (T10) of the period (T1) is further behind a lead time (L); (120) data collection: each patent entity (PE) within the first time period (T1), the patent indicator (PI) a plurality of first patent indicator data (121P) corresponding to the financial indicator (FI) and a plurality of first financial indicator data (121F), collecting each of the patent entities (PE) in the second time period (T2) Within the patent indicator (PI) and the financial indicator (FI) corresponding to a plurality of second patent indicator data (122P) and a plurality of second financial indicator data (122F); (130) the first number of patent indicators (121P), the first financial indicator data (121F), the second patent indicator data (122P), and the second financial indicator data (122F) form a first panel data (131); (140) provide a positive The state distribution conversion program (141) converts the first panel data (131) into a second panel data (142); (150) provides a first time series operation program based on the time lead (L) (151) At least one first fitting coefficient (152), and a first threshold (153) corresponding to the first fitting coefficient (152), the self-variable of the first time series operation program (151) is the first One of the patent indicators (PI) of the second panel data (142), the variable is the financial indicator (FI) of the second panel data (142); (160) the first time series operation program (151) is used successively, Computing the second panel data (142), and filtering, by the plurality of patent indicators (PI), the plurality of patent core indicators (PCI) that the first fitting coefficient (152) meets the first threshold (153); (170) providing a second time series operation program (171) based on the time lead period (L), at least one second fitting coefficient (172), and a first corresponding to the second fitting coefficient (172) a second threshold (173), the independent variable of the second time series operation program (171) is all the patent core indicators (PCI) in the second panel data (142), and the variable is the second panel data (142) The financial indicator (FI); (180) through the second time series operation program (171) and the time lead period (L), the calculation generates a patent leading equation (181), the patent leading equation (181) is composed of A patent leading indicator (PLI) and a corresponding weighting factor (182) of each of the patent leading indicators (PLI), the patent leading indicator (PLI) selected by the patent core indicator (PCI), in the patent leading equation ( 181), the second fitting coefficient (172) of each patent leading indicator (PLI) conforms to the second threshold (173); (190) introducing the first patent indicator data (121P) into the patent leading formula (181) generating a patent lead score (191) for each of the patent entities (PE); and (200) based on the leading score of the patent (191) for stock. According to the patented big data forecasting stock selection method (100) implemented by the computer according to claim 1, wherein the time lead period (L) is one quarter, and the length of time (T0) is one year. According to claim 2, the patented big data forecasting stock selection method (100) implemented by the computer, wherein the number of patent core indicators (PCI) is less than the number of the patent index (PI), the patent leading indicator The number of (PLI) is less than the number of core patents (PCI). According to claim 3, the patent big data forecasting stock selection method (100) implemented by the computer, wherein the patent index (PI) includes an index describing the invention patent, an index describing the invention patent, and a description of the utility model patent. Indicators, indicators describing design patents, and indicators describing effective patents. The patent big data forecasting stock selection method (100) implemented by the computer according to claim 4, wherein the patent indicator (PI) includes a length of time before the end date (T10) of the first time period (T1) The index of the patent within (T0) and multiple time lengths (T0). According to claim 1, the patent big data prediction stock selection method (100) implemented by the computer, wherein the first fitting coefficient (152) and the second fitting coefficient (172) are p values, the first The threshold (153) and the second threshold (173) are not greater than 0.1. A patented big data predictive stock selection method (600) implemented by a computer, comprising the following steps: (610) setting parameters: multiple patent entities (PE), one time lead period (L), one first year (T1), a second year (T2), a plurality of patent indicators (PI) and a financial indicator (FI) for describing the patent entity (PE) in the first year (T1) and the second year (T2) The patent index (PI) includes an index describing the patent for the invention, an index describing the patent granted by the invention, an indicator describing the utility model patent, an index describing the design patent, and an indicator describing the effective patent, wherein the patent entity (PE) For a listed company, the financial indicator (FI) is the reinstatement closing price (SP) of each patent entity (PE) on the last trading day of the first year (T1) and the second year (T2), The length of the first year (T1) and the second year (T2) is one year, and the end date (T20) of the second year (T2) is more than the end date (T10) of the first year (T1). Behind a lead time (L), the lead lead (L) is a quarter; (620) Collecting data: each of the patent entities (PE) In the first year (T1), the patent indicator (PI) and the plurality of first patent indicator data (121P) corresponding to the financial indicator (FI) and the plurality of first financial indicator data (121F) are collected. In the second year (T2), the patent entity (PE) has a plurality of second patent indicator data (122P) corresponding to the financial indicator (FI) and a plurality of second financial indicator data ( 122F); (630) forming the first patent indicator data (121P), the first financial indicator data (121F), the second patent indicator data (122P), and the second financial indicator data (122F) a panel data (131); (640) provides a normal distribution conversion program (141), the first panel data (131) is converted into a second panel data (142); (650) providing a one-dimensional Granger causality test model (151) based on the time lead (L), at least one first fitting coefficient (152), and a first corresponding to the first fitting coefficient (152) a threshold value (153), the independent variable of the unary Granger causality test model (151) is a patent index (PI) of the second panel data (142), and the variable is the second panel data (142) The financial indicator (FI); (660) successively using the one-dimensional Granger causality test model (151), computing the second panel data (142), and screening a plurality of patents from the plurality of patent indicators (PI) The core indicator (PCI), the first fitting coefficient (152) of each of the patent core indicators (PCI) conforms to the first threshold (153); (670) provides a multivariate Granger based on the lead time (L) a causality test model (171), at least one second fit coefficient (172), and a second threshold (173) corresponding to the second fit coefficient (172), the multivariate Granger causality test model (171) The self-variable is all the patent core indicators (PCI) in the second panel data (142), and the variable is the financial indicator (FI) of the second panel data (142); (680) Through the multivariate Granger causality test model (171), the calculus generates a patented leading equation (181), which is based on a number of patent leading indicators (PLI) and each of the patent leading indicators (PLI). The weighting factor (182) is composed of the patent leading indicator (PLI) selected by the patent core index (PCI). In the patent leading equation (181), the second fitting of the patent leading indicator (PLI) The coefficient (172) meets the second threshold (173); (690) generates a patent lead score (191) for each patent entity (PE) by the patent leading equation (181) and the first patent indicator data (121P). (700) based on the patent leading score (191) and the first financial indicator data (121F), generating a predicted value of the compound price of each of the patent entities (PE) (701); and (710) the right to reinstate The price yield forecast (701) is sorted as a stock selection basis. A patented big data predictive stock selection method (800) implemented by a computer, comprising the following steps: (810) setting parameters: multiple patent entities (PE), one time lead period (L), one first year (T1), a second year (T2) to describe each of the patent entities (PE) in the first year (T1) And a plurality of patent indicators (PI) and a financial indicator (FI) of the second year (T2), the patent indicator (PI) includes an indicator describing the invention patent, an indicator describing the patent granted by the invention, and a description of the utility model patent. Indicators, indicators describing design patents, and indicators describing effective patents, wherein the patent entity (PE) is a listed company, and the financial indicator (FI) is the patent entity (PE) in the first year (T1) With the reinstatement closing rate (SPR) of the last trading day in the second year (T2), the length of the first year (T1) and the second year (T2) is one year, the second The end of the year (T2) (T20) is one time behind the end of the first year (T1) (T10), which is the lead period (L), which is the lead period (L) for one quarter; (820) collection Data: each of the patent entities (PE) in the first year (T1), the patent indicator (PI) and the financial indicator (FI) corresponding to a plurality of first patent indicator data (121P) and a plurality of first Financial indicator data (121F), collecting each patent entity (PE) in the second year (T2), the patent indicator (PI) and the financial indicator (FI) a plurality of second patent indicator data (122P) and a plurality of second financial indicator data (122F); (830) the first patent indicator data (121P), the first financial indicator data (121F), the first The second patent indicator data (122P) and the second financial indicator data (122F) constitute a first panel data (131); (840) provide a normal distribution conversion program (141), the first panel data (131) Converting to a second panel data (142); (850) providing a one-dimensional Granger causality test model (151) based on the time lead (L), at least one first fit coefficient (152), and corresponding to a first threshold (153) of the first fitting coefficient (152), the independent variable of the unary Granger causality test model (151) being a patent indicator (PI) of the second panel data (142), The variable is the financial indicator (FI) of the second panel data (142); (860) sequentially using the unary Granger causality test model (151) to calculate the second panel data (142), screening a plurality of patent core indicators (PCI) from the plurality of patent indicators (PI), and the first fitting coefficient (152) of each of the patent core indicators (PCI) conforms to the first threshold ( 153); (870) providing a multivariate Granger causality test model (171) based on the time lead (L), at least one second fit coefficient (172), and corresponding to the second fit coefficient (172) a second threshold (173), the independent variable of the multivariate Granger causality test model (171) is all the patent core indicators (PCI) in the second panel data (142), and the variable is the second panel The financial indicator (FI) of the data (142); (880) through the multivariate Granger causality test model (171), the calculus generates a patent leading equation (181), which leads by multiple patents. The indicator (PLI) and each of the patent leading indicators (PLI) corresponding weight coefficient (182), the patent leading indicator (PLI) selected by the patent core indicator (PCI), in the patent leading equation (181) The second fitting coefficient (172) of each of the patent leading indicators (PLI) conforms to the second threshold (173); (890) passes the patent leader The program (181) and the first patent indicator data (121P) generate a patent lead score (191) of each patent entity (PE), and the patent lead score (191) is the reclosing closing price of each patent entity (PE). Yield (SPR) forecast; and (900) rank the patent lead score (191) as a stock selection basis. According to claim 8, the patent big data forecasting stock selection method (800) implemented by the computer, wherein the return rate of return (SPR) is the annual rate of return. A patented big data predictive stock selection computer system (400) for implementing the patented big data forecasting stock selection method by computer according to any one of claims 1 to 9, the patent big data predicting a computer for stock selection The system (400) includes: an indicator calculation unit (420) for calculating a patent indicator (PI) of the patent entity (PE) and the financial indicator (FI), generating the first patent indicator data (121P), The first financial indicator data (121F), the second patent indicator data (122P) and the second financial indicator data (122F); a database unit (430) for storing information of the patent entity (PE), The first patent indicator data (121P), the first financial indicator data (121F), the second patent indicator data (122P) and the second Financial indicator data (122F); a calculation and prediction unit (440) for calculating the patent core indicator (PCI), the patent leading indicator (PLI), the patent leading equation (181), and the patent leading score (191) a display and export unit (450) for presenting the patent entity (PE) and its corresponding patent lead score (191); and a core control unit (410) for integrating and manipulating the indicator calculation unit (420), the database unit (430), the calculation and prediction unit (440), and the display and export unit (450).