CN110349027A - Pairs trade system based on deeply study - Google Patents
Pairs trade system based on deeply study Download PDFInfo
- Publication number
- CN110349027A CN110349027A CN201910634385.XA CN201910634385A CN110349027A CN 110349027 A CN110349027 A CN 110349027A CN 201910634385 A CN201910634385 A CN 201910634385A CN 110349027 A CN110349027 A CN 110349027A
- Authority
- CN
- China
- Prior art keywords
- module
- assets
- actor
- pairs trade
- transaction
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
- G06Q40/00—Finance; Insurance; Tax strategies; Processing of corporate or income taxes
- G06Q40/04—Trading; Exchange, e.g. stocks, commodities, derivatives or currency exchange
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
- G06Q40/00—Finance; Insurance; Tax strategies; Processing of corporate or income taxes
- G06Q40/06—Asset management; Financial planning or analysis
Landscapes
- Business, Economics & Management (AREA)
- Engineering & Computer Science (AREA)
- Finance (AREA)
- Accounting & Taxation (AREA)
- Development Economics (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- General Business, Economics & Management (AREA)
- Economics (AREA)
- Marketing (AREA)
- Strategic Management (AREA)
- Technology Law (AREA)
- Entrepreneurship & Innovation (AREA)
- Operations Research (AREA)
- Human Resources & Organizations (AREA)
- Game Theory and Decision Science (AREA)
- Financial Or Insurance-Related Operations Such As Payment And Settlement (AREA)
Abstract
The invention discloses the pairs trade systems learnt based on deeply, specifically include with lower module: historical trading data module, obtain historical trading data from transaction platform, and pre-process to data;Co integration test module examines the basic condition of assets, does correlation analysis to the historical price sequence of similar property, selects the biggish assets of correlation to co integration test is carried out, and assists the assets of whole relationship to as trading object for meeting;Based on the pairs trade module of deeply study, analyzed for the historical transactional information to trading object, and export result;Strategy selection module selects transaction movement according to the result of the pairs trade module learnt based on deeply;Data base update module deletes state, reward and the return being stored in earliest in due course, is stored in newest state.The present invention combines the decision-making capability of the ability in feature extraction of deep learning and intensified learning, can adapt to the variation of financial market environment, can obtain fair margin of profit.
Description
Technical field
The present invention relates to pairs trade fields, and in particular to the pairs trade system based on deeply study.
Background technique
In recent years, with the rapid development of economy, the financial products such as digital cash, fund-raising gap are in succession in financial market
Upper appearance, by a large amount of positive researches, pairs trade still is able to bring to investor in the financial market of market environment complexity
Profit.In addition the fast development of internet and artificial intelligence technology also pushes the big step of traditional financial industry to advance, and utilizes meter
Due to psychological factor when the automated transaction that calculation machine technology carries out can also avoid manual transaction while reducing cost of labor
Caused by adverse effect, and be all incomparable in the quantization in high frequency transaction field and upper manual transaction of trading.High frequency is handed over
Easily and automated transaction produces a large amount of historical transactional information, and pairs trade and machine learning are combined, analyzes, excavate this
Information in a little historical trading datas, can predict future market state, provide reliable reference for investor.
Pairs trade is a simple but very important investment tactics in statistical arbitrage and quantization investment, and pairs trade is logical
Often it is divided into the progress of two steps, i.e., first selects long-term relevant transaction in assets pair, then find the chance of transaction, thus to pairs trade
Research also can substantially be divided into two aspects, and first is that long-term relevant assets pair how are picked out from numerous assets, the
Second is that how to establish Trading Model to find suitable Transaction apparatus meeting.At present mainly there is the method for selection transaction pair in pairs trade
Furthest Neighbor, random price differential method and association it is whole three kinds of method, Transaction apparatus can discovery rely primarily on traditional machine learning, intensified learning,
Deep learning and deeply study.
Summary of the invention
The pairs trade system that the purpose of the present invention is to propose to be learnt based on deeply is handed over by using the whole method selection of association
It is easily right, then deeply study is realized automatic learning characteristic and decision, avoided to a certain extent in conjunction with pairs trade
It is lost caused by subjective factors in expert system, can also obtain fair margin of profit in face of financial market complicated and changeable.
The technical solution adopted by the invention is as follows:
Based on the pairs trade system of deeply study, the module of system includes:
Historical trading data module obtains historical trading data from transaction platform, and pre-processes to data;
Co integration test module examines the basic condition of assets, does correlation analysis to the historical price sequence of similar property,
It selects the biggish assets of correlation to co integration test is carried out, assists the assets of whole relationship to as trading object for meeting;
Based on the pairs trade module of deeply study, analyzed for the historical transactional information to trading object,
And export result;
Strategy selection module selects transaction movement according to the result of the pairs trade module learnt based on deeply;
Data base update module deletes state, reward and the return being stored in earliest in due course, is stored in newest state.
Further, the correlation analysis of the co integration test module refers to that for two assets sequence variables be X=(x1,
x2..., xT) and Y=(y1, y2..., yT), then its relative coefficient R can be expressed as
Further, co integration test refers to the whole property of association that assets pair are examined by EG two-step method in the co integration test module.
Further, it is described based on deeply study pairs trade module in by Actor-Critic intensified learning with follow
Ring neural network LSTM is combined, and the plan of Actor in approximate Actor-Critic is gone using the identical neural network of two structures
Slightly function πθThe value function v of (a | s) and Criticπ(st).Specific steps are as follows:
(1) it is input to the history feature of product as ambient condition in Actor network, generates respective action;
(2) pairs trade system makes corresponding operation according to movement, returns to the profit of transaction as reward, reward is passed
Critic network is given, for assessing TD error;
(3) Critic network makes assessment to movement and the reward generated, and TD error is then transmitted to Actor network and is used for
Update Actor network parameter.
The specific frame of Critic network are as follows: first layer LSTM, the second layer are hidden layer, and the result of third layer output is made
For value function vπ(st), then according to environment to movement atThe reward r of returnt, generate time difference error (TD), calculation formula
It is as follows:
Et_error=rt+ξvπ(st)-vπ(st+1), 0≤ξ≤1
The objective function of Cirtic is to minimize error function, is denoted asWhereinMeet
Following formula:
The specific frame of Actor network are as follows: first layer LSTM, the second layer are hidden layer, and third layer is swashed using softmax
Function living generates the probability distribution value for selecting each transaction movement.The gradient updating calculation method of Actor are as follows:
In conclusion by adopting the above-described technical solution, the beneficial effects of the present invention are:
1. using Actor-Critic intensified learning method in the present invention, it can be carried out single step update, be easy to converge to part
Optimal solution has the ability for exporting optimal stable strategy.And the existing pairs trade method based on deeply learning method
It using Policy-Gradient method, is updated using single step, learning efficiency is low and variance with higher.
2. the ability in feature extraction in the present invention using LSTM avoids the artificial subjectivity for extracting feature, greatly improve
In pairs trade discovery Transaction apparatus can ability.
3. the research object of existing pairs trade is mainly the financial products such as the stock of market concentration, futures, this patent can
High in face of transaction frequency, market dispersion Selection of Financial Assets transaction movement, to earn a profit.
Detailed description of the invention
Fig. 1 system architecture diagram.
Two step co integration test method schematic diagram of Fig. 2 EG
Fig. 3 quantifies trade network structure chart based on the pairing that deeply learns
Specific embodiment
With reference to the accompanying drawings and embodiments, a specific embodiment of the invention is described further.
Based on the pairs trade system of deeply study, such as Fig. 1 includes with lower module:
Historical trading data module obtains historical trading data from transaction platform, and pre-processes to data;
Co integration test module examines the basic condition of assets, does correlation analysis to the historical price sequence of similar property,
It selects the biggish assets of correlation to co integration test is carried out, assists the assets of whole relationship to as trading object for meeting;
Based on the pairs trade module of deeply study, analyzed for the historical transactional information to trading object,
And export result;
Strategy selection module selects transaction movement according to the result of the pairs trade module learnt based on deeply;
Data base update module deletes state, reward and the return being stored in earliest in due course, is stored in newest state.
In the present embodiment, system operational process mainly passes through deeply learning method selection progress pairs trade,
As shown in Figure 1.
In historical trading data module, historical trading data is obtained from the API of transaction platform in the present embodiment, selection
Characteristic are as follows: timestamp, opening price, closing price, highest price, lowest price, newest knock-down price and trading volume.Due to acquisition time
Disunity extracted acquired data with one minute for interval in the present embodiment, so that the data volume between different transaction platforms is protected
It holds unanimously, being filled with None for intermediate data missing later again weeds out the row of shortage of data.Due to price differential and logarithm
Earning rate is smaller and sometimes positive number is sometimes negative, and the numerical value of other features is also not of uniform size, more for model
It is trained well, in the present embodiment to flat fare, opening price, closing price, highest price, lowest price and newest knock-down price, trading volume with
And Relative Strength Index is normalized, circular is as follows:
For t moment characteristic sequence Xt=(x1, x2..., xn) in some feature xi, minimax normalizing first is carried out to it
Change, then by xiIt zooms on section [0,1], calculation formula is as follows:
Again by normalized result xi_01It is mapped on section [- 1,1], calculation formula is as follows:
Here xi_minIndicate feature x in training dataiMinimum value, xi_maxIndicate feature x in training dataiMaximum
Value.
Use that specific step is as follows in co integration test module, in the present embodiment:
1. passing through analysis relevant financial transaction in assets amount size, liquidity of assets, position in industry and the technology of behind back
The basic conditions such as scape pick out similar assets as a set C;
2. assets are matched two-by-two in couple C, correlation analysis is done to its historical price sequence, if two assets sequence variables are
X=(x1, x2..., xT) and Y=(y1, y2..., yT), then its relative coefficient R can be indicated are as follows:
HereWithVariable X and the mean value of Y are respectively indicated, the range of R is between [- 1,1], when the absolute value of R | R | more
Greatly, correlation is stronger between showing two variables.It is selected later according to relative coefficient from big to small to assets to sequence
The higher k of related coefficient ranking is to assets as pre-selection assets to C_pre.
3. examining the whole property of association of assets pair in C_pre using EG two-step method, as shown in Figure 2.It is arbitrarily chosen from C_pre first
Select a pair of of assets Ci=1,2 ..., k first judges CiIn the historical price sequences of each assets whether be that single order list is whole, if
It is that regression model y is done to the historical price sequence of assets pairt=alpha+beta xt+εt, recycle ADF test statistics method to examine residual
Poor εtStationarity, if residual error is smoothly, to illustrate that it has and assist whole relationship, if unstable, by CiFrom C_pre
It rejects, continues to examine the whole property of association of lower a pair of of assets in C_pre.Then the assets in C_pre all meet the whole relationship of association, are made
For research object, it is updated in the pairing quantization transaction system based on deeply study and carries out pairs trade.
In the pairs trade module learnt based on deeply, such as Fig. 3, deeply study in the present embodiment be by
Actor-Critic intensified learning is combined with Recognition with Recurrent Neural Network LSTM, goes approximation using the identical neural network of two structures
The strategic function π of Actor in Actor-CriticθThe value function v of (a | s) and Criticπ(st).In the present embodiment, specifically
Steps are as follows:
(1) it is input to the history feature of product as ambient condition in Actor network, generates corresponding movement;
(2) pairs trade system makes corresponding operation according to movement, returns to the profit of transaction as reward, reward is passed
Critic network is given, for assessing TD error;
(3) Critic network makes assessment to movement and the reward generated, and TD error is then transmitted to Actor network and is used for
Update Actor network parameter.
The specific frame of Critic network are as follows: first layer LSTM, the second layer are hidden layer, and the result of third layer output is made
For value function vπ(st), then according to environment to movement atThe reward r of returnt, generate time difference error (TD), calculation formula
It is as follows:
Et_error=rt+ξvπ(st)-vπ(st+1), 0≤ξ≤1
The objective function of Cirtic is to minimize error function, is denoted asWhereinMeet
Following formula:
The specific frame of Actor network are as follows: first layer LSTM, the second layer are hidden layer, and third layer is swashed using softmax
Function living generates the probability distribution value for selecting each transaction movement.The gradient updating calculation method of Actor are as follows:
The pairs trade method used in the present embodiment is as follows:
Motion space is denoted asIt respectively indicates and buys in, sells, holds and stop loss (only
It is full of).When Actor network selection acts atTo indicate to buy in A product when buy, B product is sold;Work as atTo sell A product when sell,
Buy in B product;Work as atTo keep A product and B product constant when hold;Work as atThen to close a position processing when stop i.e. while A production of selling short
Product and B product.Platform can collect the formality expense of a part when transaction, and rate is denoted as fee, therefore the profit of t moment transaction, i.e.,
The reward r that environment generates in deeply studytThe total assets profit of product after service charge can be deducted with transactiontSubtract friendship
Total assets profit before easilyt-1Obtain rt=profitt-profitt-1.A product and B product after transaction are utilized when calculating total assets
Remaining quantity A_numst、B_numstMultiplied by corresponding price A_Pt, B_Pt, it may be assumed that
profitt=A_numst×A_Pt+B_numst×B_Pt.
If the trading volume of A product and B product is respectively A_vol in transactiont、B_volt, then according to different movement atProduct
Surplus A_numst、B_numstChange as follows:
1. working as atWhen for buy, A product is bought in, B product is sold.Formality rate product surplus in transaction is deducted to be respectively as follows:
2. working as atWhen for sell, A product is sold, the corresponding surplus of B product is bought in and is respectively as follows:
B_numst=B_numst-1+B_volt×(1-fee);
3. working as atWhen for hold, A product and B product are not traded, trading volume 0.Therefore:
B_numst=B_numst-1;
4. working as atWhen for stop, while sell short A product and B product, surplus is 0 at this time, total assets are as follows:
In the pairs trade module learnt based on deeply, ambient condition stIt is with the history for assisting whole relationship product
Time series feature, strategic function πθ(a | s) it is obtained using Actor network approximation, value function vπ(s) it is gone by Critic network
Approximation acts atIt is input in pairs trade method, the profit that method obtains is as corresponding reward rt.Final goal is maximum
Change accumulative return R, wherein 0≤ξ≤1 is discount factor.
The data set division methods of model training are as follows in the present embodiment:
Data set is divided according to the form of training period transaction period, i.e., makees from the data that most start time chooses T time
For training data, the data of T_test time are and then chosen after T time as transaction data, later reselection T time
Data are recycled according to this as training until trained and all data of having traded.
In strategy selection module, the motion space for including in the present embodiment is Root
Transaction movement is selected according to the output of deeply study module, as transaction movement atWhen for buy, buys in assets A and sell assets B's
The position in storehouse of B is first examined in operation, if the position in storehouse of B is not zero, i.e. investor possesses B assets, and at this moment selection buys A and sells B, if B
Position in storehouse is zero, indicates that this moment investor has not had B assets, can not sell, then A is bought in selection.Work as atWhen for sell, first examine
The position in storehouse that certification of registered capital produces A sells A and buys B, otherwise only buy in assets B if position in storehouse is not zero.Work as atWhen for hold, two kinds of moneys are kept
The position in storehouse of production is constant.Work as atWhen for stop, two kinds of assets are all sold short, position in storehouse becomes 0 entirely.Environment is returned according to the difference of movement
It returns different return and rewards rt。
In data base update module, electing sale movement returns to reward r in strategy selection moduletAfterwards, by state stWith
st+1With movement atAnd reward rtIt is stored in data base R, compares transaction count n and data base size RNIf n is greater than RN, then
Updating data base is to delete the state, reward and the return that most start deposit, and newest state is deposited into R, can be protected in this way
Card sampling is the information of nearest time when updating model parameter.
The invention discloses the pairs trade system learnt based on deeply, all made on the basis of the present invention
What modifications, equivalent substitutions and improvements etc., as long as principle is identical, should all be included in the protection scope of the present invention.
Claims (4)
1. the pairs trade system based on deeply study characterized by comprising
Historical trading data module obtains historical trading data from transaction platform, and pre-processes to data;
Co integration test module examines the basic condition of assets, does correlation analysis to the historical price sequence of similar property, selection
The biggish assets of correlation assist the assets of whole relationship to as trading object to co integration test is carried out, by meeting;
Based on the pairs trade module of deeply study, analyzed for the historical transactional information to trading object, and defeated
Result out;
Strategy selection module selects transaction movement according to the result of the pairs trade module learnt based on deeply;
Data base update module deletes state, reward and the return being stored in earliest in due course, is stored in newest state.
2. the pairs trade system according to claim 1 based on deeply study, it is characterised in that: the whole inspection of association
The correlation analysis for testing module refers to that for two assets sequence variables be X=(x1, x2..., xT) and Y=(y1, y2..., yT),
Then its relative coefficient R can be indicated are as follows:
3. the pairs trade system according to claim 1 based on deeply study, it is characterised in that: the whole inspection of association
It tests co integration test in module and refers to the whole property of association for examining assets pair by EG two-step method.
4. the pairs trade system according to claim 1 based on deeply study, it is characterised in that: described based on deep
It spends in the pairs trade module of intensified learning and combines Actor-Critic intensified learning with Recognition with Recurrent Neural Network LSTM, use
The identical neural network of two structures removes the strategic function π of Actor in approximate Actor-CriticθThe value of (a | s) and Critic
Function vπ(st).Specific steps are as follows:
(1) it is input to the history feature of product as ambient condition in Actor network, generates respective action;
(2) pairs trade system makes corresponding operation according to movement, returns to the profit of transaction as reward, reward is transmitted to
Critic network, for assessing TD error;
(3) Critic network makes assessment to movement and the reward generated, and TD error is then transmitted to Actor network for updating
Actor network parameter;
The specific frame of Critic network are as follows: first layer LSTM, the second layer are hidden layer, and the result of third layer output is as value
Function vπ(st), then according to environment to movement atThe reward r of returnt, it generates time difference error (TD), calculation formula is as follows:
Et_error=rt+ξvπ(st)-vπ(st+1), 0≤ξ≤1
The objective function of Cirtic is to minimize error function, is denoted asWhereinMeet following formula:
The specific frame of Actor network are as follows: first layer LSTM, the second layer are hidden layer, and third layer activates letter using softmax
Number generates the probability distribution value for selecting each transaction movement.The gradient updating calculation method of Actor are as follows:
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201910634385.XA CN110349027A (en) | 2019-07-19 | 2019-07-19 | Pairs trade system based on deeply study |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201910634385.XA CN110349027A (en) | 2019-07-19 | 2019-07-19 | Pairs trade system based on deeply study |
Publications (1)
Publication Number | Publication Date |
---|---|
CN110349027A true CN110349027A (en) | 2019-10-18 |
Family
ID=68175252
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201910634385.XA Pending CN110349027A (en) | 2019-07-19 | 2019-07-19 | Pairs trade system based on deeply study |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN110349027A (en) |
Cited By (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN111325624A (en) * | 2020-02-11 | 2020-06-23 | 支付宝(杭州)信息技术有限公司 | Real-time prevention and control system and method for network transaction |
CN112364077A (en) * | 2020-11-09 | 2021-02-12 | 光大理财有限责任公司 | Training sample generation method, machine learning model training method and related device |
CN112419064A (en) * | 2020-12-07 | 2021-02-26 | 中山大学 | Energy transaction method, device and equipment based on deep reinforcement learning and alliance chain |
CN112862620A (en) * | 2021-03-31 | 2021-05-28 | 山东大学 | Investment product combination recommendation method and system based on investor preference |
CN116824207A (en) * | 2023-04-27 | 2023-09-29 | 国科赛赋河北医药技术有限公司 | Multidimensional pathological image classification and early warning method based on reinforcement learning mode |
-
2019
- 2019-07-19 CN CN201910634385.XA patent/CN110349027A/en active Pending
Cited By (8)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN111325624A (en) * | 2020-02-11 | 2020-06-23 | 支付宝(杭州)信息技术有限公司 | Real-time prevention and control system and method for network transaction |
CN111325624B (en) * | 2020-02-11 | 2022-04-26 | 支付宝(杭州)信息技术有限公司 | Real-time prevention and control system and method for network transaction |
CN112364077A (en) * | 2020-11-09 | 2021-02-12 | 光大理财有限责任公司 | Training sample generation method, machine learning model training method and related device |
CN112419064A (en) * | 2020-12-07 | 2021-02-26 | 中山大学 | Energy transaction method, device and equipment based on deep reinforcement learning and alliance chain |
CN112419064B (en) * | 2020-12-07 | 2022-02-08 | 中山大学 | Energy transaction method, device and equipment based on deep reinforcement learning and alliance chain |
CN112862620A (en) * | 2021-03-31 | 2021-05-28 | 山东大学 | Investment product combination recommendation method and system based on investor preference |
CN116824207A (en) * | 2023-04-27 | 2023-09-29 | 国科赛赋河北医药技术有限公司 | Multidimensional pathological image classification and early warning method based on reinforcement learning mode |
CN116824207B (en) * | 2023-04-27 | 2024-04-12 | 国科赛赋河北医药技术有限公司 | Multidimensional pathological image classification and early warning method based on reinforcement learning mode |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN110349027A (en) | Pairs trade system based on deeply study | |
Cheung et al. | A multi-perspective knowledge-based system for customer service management | |
US7664693B1 (en) | Financial methodology for the examination and explanation of spread between analyst targets and market share prices | |
CN109785003A (en) | A kind of Pharmaceutical retail industry medicine sales forecasting system and method | |
AU2585401A (en) | Rapid valuation of portfolios of assets such as financial instruments | |
KR20010103784A (en) | Valuation prediction models in situations with missing inputs | |
CN105259816A (en) | Bank dynamic credit monitoring system based on wireless sensing network and method | |
CN108734220A (en) | Adaptive Financial Time Series Forecasting method based on k lines cluster and intensified learning | |
Chen et al. | Short-term water demand forecast based on automatic feature extraction by one-dimensional convolution | |
CN109492838A (en) | A kind of stock index price expectation method based on deep-cycle neural network | |
Brueller et al. | Knowing when to acquire: The case of multinational technology firms | |
CN110309508A (en) | A kind of VWAP quantization transaction system and method based on investor sentiment | |
Li et al. | Learning to trade with deep actor critic methods | |
CN104346698A (en) | Catering member big data analysis and checking system based on cloud computing and data mining | |
CN112132438A (en) | Investment and financing docking and equity value evaluation system for regional equity market | |
CN117575171A (en) | Grain situation intelligent evaluation system based on data analysis | |
CN111292188A (en) | Intelligent integrated return test system for pairing transaction | |
CN112418575A (en) | Futures initiative contract quantitative timing decision system based on cloud computing and artificial intelligence deep learning algorithm | |
Balcarová | Is competitiveness in the European Union converging | |
CN113449923A (en) | Multi-model object market quotation prediction method and device | |
CN111222914A (en) | Accurate auxiliary type delivery system for goods based on e-commerce platform | |
Moody | Forecasting the economy with neural nets: A survey of challenges and solutions | |
Kangane et al. | Analysis of different regression models for real estate price prediction | |
Bahremand et al. | Providing Financial Flow Management Strategies in Supply Chain Projects | |
Sun et al. | Credit Risk Simulation of Enterprise Financial Management Based on Machine Learning Algorithm |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
PB01 | Publication | ||
PB01 | Publication | ||
SE01 | Entry into force of request for substantive examination | ||
SE01 | Entry into force of request for substantive examination | ||
WD01 | Invention patent application deemed withdrawn after publication | ||
WD01 | Invention patent application deemed withdrawn after publication |
Application publication date: 20191018 |