CN103279805B - A kind of stock certificate data analytical approach based on price linkage network - Google Patents

Abstract

The invention discloses a kind of stock certificate data analytical approach based on price linkage network, the steps include: 1) collect stock price data, calculate the continuous action relation between stock price, take stock as node, continuous action relation builds price linkage network for limit; 2) in price linkage network, the appreciation according to the weight calculation stock node of father node collection, stock price tendency and continuous action relation within the double bounce of stock node is expected; 3) expect to sort to stock according to appreciation.The inventive method fully excavates the continuous action relation between stock price, can rationally judge that the appreciation of each stock is at no distant date expected, effectively avoid the fair game problem that may run in single stock price forward prediction according to the price volalility situation of stock market.The inventive method calculates simple, has ageing, dirigibility and extendability, little to stock historical data treatment capacity, is applicable to the large and price volalility feature frequently of stock market data amount.

Description

A kind of stock certificate data analytical approach based on price linkage network

Technical field

The present invention relates to a kind of data analysing method, specifically a kind of stock certificate data analytical approach based on price linkage network, for predicting stock price tendency.

Background technology

Investment in stocks needs to make reasonable prediction to the direction of stock future trend and possibility.In stock market, the factor that can affect share price is varied, comprises economic environment, national policy, buyer's psychology etc., makes the feature of stock price tendency be difficult to accurate assurance.Stock exchange have accumulated a large amount of historical data, and current method carries out treatment and analysis to historical data, the association between mining data, finds out Changing Pattern and founding mathematical models, predict on this basis to stock price tendency.

Existing stock price data Forecasting Methodology can be divided into two large classifications: a class is that the list based on statistical theory props up Stock Price Fluctuation Forecasting Methodology; Another kind of is correlation rule Forecasting Methodology based on the study such as neural network, fractal theory, support vector machine and digging technology.Price volalility Forecasting Methodology, based on the time series data of stock price, is attempted finding trend wherein and rule, is predicted the future trend of stock; It will reappear in future according to the historical development trend being stock price.The Forecasting Methodology of correlation rule mainly studies correlativity between stock price and trading volume, between different classes of stock, between stock and other types commodity or cause-effect relationship, the state current according to associated articles, according to the future trend of the prediction stocks such as cause-effect relationship; Its foundation is that the correlation rule occurred in history can reappear in the Stock Price Fluctuation in future.

If stock price to be defined as a stochastic variable, then stock price volalility process in the market can be considered as a stochastic process.Because stock price is by various factors, Stock Trading is equivalent to the game between buyer and seller.Along with the raising of both parties' information symmetrical, a fair game (being defined as halter strap) is tended in Stock Trading, it is characterized in that the value of stock price subsequent period is expected to equal the last observed reading.This point is similar to coin and throws, and all faces up for 99 times even if throw in history, and when throwing for the 100th time, heads expected probability still equals the probability thrown for the 99th time.If there is fair game, then only according to the historical price sequence of single stock, be difficult to the future trend accurately estimating stock.Stock certificate data Forecasting Methodology at present based on correlation rule obtains increasing investigation and application, but existing method computation complexity is too high, lacks necessary screening to historical data; Thus timely response is difficult to fast changing stock market, be not suitable for the Prediction of Stock Index requirement of short-term investment.

Summary of the invention

Technical matters to be solved by this invention be according to stock price historical data between continuous action relation, analysis list stock share price data tendency at no distant date accurately.

A kind of stock certificate data analytical approach based on price linkage network of the present invention, it comprises the following steps:

1) first collect the price volalility data of stock market, calculate the price linkage relation between each stock, build price linkage network;

2) in price linkage network, the appreciation according to the weight calculation stock node of the father node collection within stock price node double bounce, stock price tendency and continuous action relation is expected;

3) expect to sort to stock price tendency according to appreciation.

The process of described step 1) is: the price data first obtaining stock market all stock at no distant date; Then to each stock, the price vector of stock is organized in chronological order; Following setting-up time is poor, respectively according to the price vector of stock last period and a rear period, calculates the price linkage relation weight of any two stocks; For continuous action relation weight setting threshold value w, retain the continuous action relation that weight is not less than threshold value w; Last is node with stock, continuous action relation builds price linkage network for directed edge, is the digraph of Weight on a limit.

The processing procedure calculating the price linkage relation weight of any two stocks in described step 1) is: first seclected time interval, can be set as hour; Given stock s _i, obtain continuous l hour s _iknockdown price, composition s _iprice vector ξ _i=<x _{i, 1}, x _{i, 2}..., x _i,l>; Element in vector is the knockdown price of sampling, and l is vector length; Then setting-up time difference t, equally by hour in units of; Next to two stock s _iand s _j, according to the price vector of both Different periods, adopt following formulae discovery from s _ito s _jcontinuous action relation weight w _{<i, j>}:

$w_{<i,j>}\frac{{\mathrm{\&Sigma;}}_{k=1}^l(x_{i,k}-{\stackrel{\&OverBar;}{x}}_i)(x_{j,k}-{\stackrel{\&OverBar;}{x}}_j)}{(l-1)\&CenterDot;{\mathrm{\&sigma;}}_i\&CenterDot;{\mathrm{\&sigma;}}_j}$

Wherein x _i,kand x _j,kbe respectively the element in price vector; s _iprice vector ξ _i=<x _{i, 1}, x _{i, 2}..., x _i,l> and s _jprice vector ξ _j=<x _{j, 1}, x _{j, 2}..., x _j,l> takes from the different periods, s _jfirst price element x _{j, 1}sampling time compare s _ifirst price element x _{i, 1}evening, mistiming t, reflected stock s _iprice volalility to stock s _jthe degree had an impact; with represent price vector ξ respectively _iand ξ _javerage, σ _iand σ _jthen represent vectorial standard deviation respectively; With stock s _iprice vector be example, and σ _icomputing formula as follows:

${\stackrel{\&OverBar;}{x}}_i=\frac{1}{l}{\mathrm{\&Sigma;}}_{k=1}^l{x}_{i,k}$

${\mathrm{\&sigma;}}_i=\sqrt{\frac{1}{l-1}{\mathrm{\&Sigma;}}_{k=1}^l{(x_{i,k}-{\stackrel{\&OverBar;}{x}}_i)}^2}$

Continuous action relation between stock is oriented, from stock s _ito stock s _jcontinuous action relation <s _i, s _j> is different from from stock s _jto stock s _icontinuous action relation <s _j, s _i>; What both adopted when calculating weight is different price vectors.

Described parameter l and t adjusts according to the price volalility situation of stock market, and when market value of shares fluctuation is violent, the short-term investment time cycle, l and t value reduced, otherwise then can increase in short-term.

Above-mentioned steps 2) in calculate shares changing tendency mark treatment scheme be: tendency is marked with three values :-1, represents price continuous decrease; 0, represent price maintenance and stablize; 1, represent price and continue to rise; For calculating stock s _itendency mark r _i, based on stock s _iprice vector ξ _i=<x _{i, 1}, x _{i, 2}..., x _i,l>, in vector, element is stock price, and parameter l arranges same step 1), is one group, is divided into l/n group with wherein continuous print n knockdown price; Each group calculating mean value, in chronological order, determines r by l/n value point _ivalue, point 3 kinds of situations:

Situation 1. is worth point and continues to rise, or is only no more than 10% drop point of parameter l, and drop range is no more than 2%, now r _i=1;

Situation 2. is worth some continuous decrease, or is only no more than 10% rising point of parameter l, and amount of increase is no more than 2%, now r _i=-1;

Situation 3. does not meet other all scenarios of situation 1 or 2, now r _i=0.

Above-mentioned steps 2) in according in double bounce father node collection calculate appreciation in share value expect treatment scheme be: given stock node s _a, first consider s _afather node collection wherein each node has sensing s _acontinuous action relation, i.e. directed edge; To set in each node s _i, calculate s _ito s _aappreciation affect e _{<i, a>}, equal directed edge <s _i, s _athe weight w of > _{<i, a>}; Then s is considered _adouble bounce father node collection node wherein needs just can be connected to s through two directed edges _a; To set in each node s _j, calculate s _jto s _aappreciation affect e _{<j, a>}, be s _jto s _aall double bounce paths on, comprise 2 continuous action relations, the maximal value of limit weight product; On this basis, stock s is calculated _aappreciation expect δ _a, formula is as follows:

${\mathrm{\&delta;}}_a=\underset{S_i\&Element;{\mathrm{\&Gamma;}}_a^1\&cup;{\mathrm{\&Gamma;}}_a^2}{\mathrm{\&Sigma;}}{r}_i\&CenterDot;e_{<i,a>}$

Wherein r _ifor stock s _itendency mark

The present invention is based on the price linkage relation between stock, by building price linkage network, topology Network Based, continuous action relation weight and association stock price tendency, the appreciation that estimate sheet props up stock is expected.Application the inventive method according to the price volalility situation of stock market, can be excavated and utilize the association between stock and influence each other; Effectively avoid the fair game problem that may run in single stock price forward prediction.The inventive method calculates simple, there is ageing, dirigibility and extendability, few to the treatment capacity of stock market historical price data, be applicable to the large and price volalility feature frequently of stock market data amount, the stock certificate data that can realize following one period fast, accurately and is in time looked forward to.

Accompanying drawing explanation

Fig. 1 is the overall framework of the stock recommend method based on price linkage network;

Fig. 2 is the treatment scheme analyzed stock price continuous action relation and build price linkage network;

Fig. 3 is the example building price linkage network according to 8 stock certificate datas;

Fig. 4 is the treatment scheme calculating stock node appreciation expectation according to price linkage network;

Fig. 5 is the example that the appreciation of accounting price interlock nodes is expected.

Embodiment

Figure 1 shows that the general technological system of the stock recommend method based on price linkage network.The input of method is the price data of stock market all stock at no distant date, and the output of method is appreciation expectation and the associated recommendation of each stock at no distant date.The inventive method comprises three modules: first according to stock price data, calculates the continuous action relation between stock price, take stock as node, continuous action relation builds price linkage network for limit; Then, in price linkage network, the appreciation according to the weight calculation stock node of father node collection, stock price tendency and continuous action relation in the double bounce of stock node is expected.

First module of the inventive method builds price linkage network, and implementation as shown in Figure 2.First the price data of stock market all stock is at no distant date obtained; Then to each stock, the price vector of stock is organized in chronological order; Following setting-up time is poor, according to the price vector of stock last period and a rear period, calculates the price linkage relation weight of two stocks; For continuous action relation weight setting threshold value, retain the continuous action relation that weight is not less than threshold value; Last is node with stock, continuous action relation builds price linkage network for directed edge, is the digraph of Weight on a limit.

For calculating the price linkage relation weight of two stocks, first seclected time interval, can be set as hour; Given stock s _i, obtain continuous l hour s _iknockdown price (getting rid of the nontransaction time), composition s _iprice vector ξ _i=<x _{i, 1}, x _{i, 2}..., x _i,l>; Element in vector is the knockdown price of sampling, and l is vector length, can be set as l=15.Then setting-up time difference t, equally by hour in units of, can t=3 be set as.Parameter l and t can adjust according to the price volalility situation of stock market, and the inventive method provides reference value; General market value of shares of working as fluctuates violent, and the short-term investment time cycle, l and t value reduced, otherwise then can increase in short-term.Next to two stock s _iand s _j, according to the price vector of both Different periods, adopt crossing dependency to calculate from s _ito s _jcontinuous action relation weight w _{<i, j>}, as shown in Equation (1):

$w_{<i,j>}=\frac{{\mathrm{\&Sigma;}}_{k=1}^l(x_{i,k}-{\stackrel{\&OverBar;}{x}}_i)(x_{j,k}-{\stackrel{\&OverBar;}{x}}_j)}{(l-1)\&CenterDot;{\mathrm{\&sigma;}}_i\&CenterDot;{\mathrm{\&sigma;}}_j}---\left(1\right)$

Wherein s _iprice vector ξ _i=<x _{i, 1}, x _{i, 2}..., x _i,l> and s _jprice vector

ξ _j=<x _{j, 1}, x _{j, 2}..., x _j,l> takes from the different periods, s _jfirst price element x _{j, 1}sampling time compare s _ifirst price element x _{i, 1}in t time interval in evening (namely hour), reflect stock s _iprice volalility to stock s _jthe degree had an impact; with represent price vector ξ respectively _iand ξ _javerage, σ _iand σ _jthen represent vectorial standard deviation respectively; With stock s _iprice vector be example, and σ _icalculating as shown in Equation (2):

${\stackrel{\&OverBar;}{x}}_i=\frac{1}{l}{\mathrm{\&Sigma;}}_{k=1}^l{x}_{i,k}$ （2）

${\mathrm{\&sigma;}}_i=\sqrt{\frac{1}{l-1}{\mathrm{\&Sigma;}}_{k=1}^l{(x_{i,k}-{\stackrel{\&OverBar;}{x}}_i)}^2}$

Continuous action relation between stock is oriented, from stock s _ito stock s _jcontinuous action relation <s _i, s _j> is different from from stock s _jto stock s _icontinuous action relation <s _j, s _i>; What both adopted when calculating weight is different price vectors.

According to computing formula, the span of continuous action relation weight is-1≤w _{<i, j>}≤ 1; Finally setting the threshold value w of continuous action relation weight, can w=0.5 be set as, retaining the continuous action relation that weight is not less than threshold value w, for building price linkage network.

The knockdown price that Fig. 3 samples within a period of time for 8 stocks, explains the building process of price linkage network.

Table a

Table b

The stock price data of table shown in a derives from Yahoo's finance and economics, reflects the price volalility of each stock in 18 hours.For building price linkage network, each setting parameter is: price vector length l=15, mistiming t=3, weight threshold w=0.5; Table b is depicted as the calculated value of continuous action relation weight between stock between two, and wherein row represents the initial stock of continuous action relation, and row represent the target stock of continuous action relation.With stock s ₁to stock s ₂continuous action relation be example, s ₁price vector be made up of 15 knockdown price data of 1 ~ 15 hour, and s ₂price vector be made up of 15 knockdown price data of 4 ~ 18 hours; Weight w is calculated according to formula (1) and (2) _<1,2>, result is 0.677; In like manner calculate stock s ₂to stock s ₁continuous action relation weight w _<2,1>, result is-0.194, and obviously both are not identical.Based on weight threshold, remove the continuous action relation of weight deficiency, constructed price linkage network as shown in Figure 3.

Second module of the present invention calculates the appreciation of often propping up stock to expect, implementation as shown in Figure 4.The first step obtains the price vector of each stock nearest period, for judging stock price tendency; Price trend marks r by tendency and represents there are three values :-1, represent price continuous decrease; 0, represent price maintenance and stablize; 1, represent price and continue to rise; The tendency value of all stock is calculated according to price vector.Next according to price linkage network, to each stock s _a, calculate its expectation of appreciating: first consider s _afather node collection wherein each node has sensing s _acontinuous action relation (directed edge); To set in each node s _i, calculate s _ito s _aappreciation affect e _{<i, a>}, equal directed edge <s _i, s _athe weight w of > _{<i, a>}; Then s is considered _adouble bounce father node collection node wherein needs just can be connected to s through two directed edges (continuous action relation) _a; To set in each node s _j, calculate s _jto s _aappreciation affect e _{<j, a>}, be s _jto s _aall double bounce paths (comprising 2 continuous action relations) on, the maximal value of limit weight product.On this basis, stock s is calculated _aappreciation expect δ _a, as shown in Equation (3):

${\mathrm{\&delta;}}_a=\underset{S_i\&Element;{\mathrm{\&Gamma;}}_a^1\&cup;{\mathrm{\&Gamma;}}_a^2}{\mathrm{\&Sigma;}}{r}_i\&CenterDot;e_{<i,a>}---\left(3\right)$

Wherein r _ifor stock s _itendency mark.As previously mentioned, r is calculated _ibased on stock s _iprice vector.Given ξ _i=<x _{i, 1}, x _{i, 2}..., x _i,l>(parameter l arranges same module 1, is generally l=15), with wherein continuously n knockdown price be one group, be divided into l/n group; Such as n=3, so be divided into 5 groups; Each group calculating mean value, obtains altogether 5 value points; In chronological order, r is determined by these 5 value points _ivalue (if parameter l value is not equal to 15, n value be not equal to 3, then a value point quantity corresponding adjustment), point 3 kinds of situations:

Situation 1. is worth point and continues to rise, or only has 1 (or being no more than 10% of parameter l) drop point, and drop range is no more than 2%, now r _i=1;

Situation 2. is worth some continuous decrease, or only has 1 (or being no more than 10% of parameter l) rising point, and amount of increase is no more than 2%, now r _i=-1;

Situation 3. does not meet other all scenarios of situation 1 or 2, now r _i=0.

Figure 5 shows that the example calculating stock node appreciation expectation in a price linkage network.Destination node is made to be s ₁, according to Fig. 5, node s ₁father node collection in comprise 3 node s ₂~ s ₄, double bounce father node collection in comprise 3 node s ₅~ s ₇.Except s ₁outward, the tendency mark of other all nodes calculates: the node tendency of its mid-ash is labeled as 0; The node tendency of band "+" is labeled as 1; The node tendency of band "-" is labeled as-1.In set in, node s ₃and s ₄to s ₁appreciation impact be respectively: e _<3,1>=0.8, e _<4,1>=0.5; In set in, from node s ₅to s ₁there are two double bounce paths: respectively through node s ₂with node s ₄, the former weight product is 0.6 × 0.6=0.36, and the latter is 0.8 × 0.5=0.4; Get maximal value wherein, therefore e _<5,1>=0.4.Do not need to consider that tendency is labeled as the node of 0, node s ₁appreciation expect be: δ ₁=0.5+0.4-0.8=0.1.

3rd module of the present invention expects to sort according to the appreciation of stock, accurate analysis and prediction single stock price trend at no distant date.The stock that expectation of appreciating is greater than 0, its price will go up; And expect that its price of stock being less than 0 will drop; The amplitude gone up or drop according to the average amplitude of association balloon/drop, can be estimated in conjunction with appreciation expectation.

The inventive method is based on the price linkage relation between stock, and by building price linkage network, the price trend of topology Network Based, continuous action relation weight and association stock, the appreciation that estimate sheet props up stock is expected, completes stock according to this and recommends.Application the inventive method according to the price volalility situation of stock market, fully can be excavated and utilizes the association between stock and influence each other, and choose reasonable is appreciated and expected that the highest stock is bought in, and recommends to appreciate simultaneously and expects that too low stock is sold; Effectively avoid the fair game problem that may run in single stock price trend prediction, improve the prospective earnings of Stock Purchase person's short-term investment.The inventive method calculates simple, there is ageing, dirigibility and extendability, few to the treatment capacity of stock market historical price data, be applicable to the large and price volalility feature frequently of stock market data amount, the analysis and prediction of stock price tendency can be completed fast, accurately and in time.

Claims (4)

1., based on a stock certificate data analytical approach for price linkage network, it is characterized in that comprising the following steps:

1) first collect the price volalility data of stock market, calculate the price linkage relation between each stock, build price linkage network; Its process is: the price data first obtaining stock market all stock at no distant date; Then to each stock, the price vector of stock is organized in chronological order; Following setting-up time is poor, respectively according to the price vector of stock last period and a rear period, calculates the price linkage relation weight of any two stocks; For continuous action relation weight setting threshold value w, retain the continuous action relation that weight is not less than threshold value w; Last is node with stock, continuous action relation builds price linkage network for directed edge, is the digraph of Weight on a limit;

Step 1) in calculate any two stocks the processing procedure of price linkage relation weight be: first seclected time interval, can be set as hour; Given stock s _i, obtain continuous l hour s _iknockdown price, composition s _iprice vector ξ _i=< x _{i, 1}, x _{i, 2}..., x _i,l>; Element in vector is the knockdown price of sampling, and l is vector length; Then setting-up time difference t, equally by hour in units of; Next to two stock s _iand s _j, according to the price vector of both Different periods, adopt following formulae discovery from s _ito s _jcontinuous action relation weight w _{< i, j >}:

$w_{<i,j>}=\frac{{\mathrm{\&Sigma;}}_{k=1}^l(x_{i,k}-{\stackrel{\&OverBar;}{x}}_i)(x_{j,k}-{\stackrel{\&OverBar;}{x}}_j)}{(l-1)\&CenterDot;{\mathrm{\&sigma;}}_i\&CenterDot;{\mathrm{\&sigma;}}_j}$

Wherein x _i,kand x _j,kbe respectively the element in price vector; s _iprice vector ξ _i=< x _{i, 1}, x _{i, 2}..., x _i,l> and s _jprice vector ξ _j=< x _{j, 1}, x _{j, 2}..., x _j,l> takes from the different periods, s _jfirst price element x _{j, 1}sampling time compare s _ifirst price element x _{i, 1}evening, mistiming t, reflected stock s _iprice volalility to stock s _jthe degree had an impact; with represent price vector ξ respectively _iand ξ _javerage, σ _iand σ _jthen represent vectorial standard deviation respectively; With stock s _iprice vector be example, and σ _icomputing formula as follows:

${\stackrel{\&OverBar;}{x}}_i=\frac{1}{l}{\mathrm{\&Sigma;}}_{k=1}^l{x}_{i,k}$

${\mathrm{\&sigma;}}_i=\sqrt{\frac{1}{l-1}{\mathrm{\&Sigma;}}_{k=1}^l{(x_{i,k}-{\stackrel{\&OverBar;}{x}}_i)}^2}$

Continuous action relation between stock is oriented, from stock s _ito stock s _jcontinuous action relation <s _i, s _j> is different from from stock s _jto stock s _icontinuous action relation <s _j, s _i>; What both adopted when calculating weight is different price vectors;

2) in price linkage network, appreciation according to the weight calculation stock node of the father node collection within stock price node double bounce, stock price tendency and continuous action relation is expected, concrete steps are, the first step obtains the price vector of each stock nearest period, for judging stock price tendency; Price trend marks r by tendency and represents; Next according to price linkage network, to each stock s _a, calculate its expectation of appreciating;

3) expect to sort to stock price data tendency according to appreciation.

2. the stock certificate data analytical approach based on price linkage network according to claim 1, it is characterized in that, described parameter l and t adjusts according to the price volalility situation of stock market, when market value of shares fluctuation is violent, the short-term investment time cycle in short-term, l and t value reduces, otherwise then can increase.

3. the stock certificate data analytical approach based on price linkage network according to claim 2, is characterized in that, above-mentioned steps 2) in calculate shares changing tendency mark treatment scheme be: tendency is marked with three values :-1, represents price continuous decrease; 0, represent price maintenance and stablize; 1, represent price and continue to rise; For calculating stock s _itendency mark r _i, based on stock s _iprice vector ξ _i=< x _{i, 1}, x _{i, 2}..., x _i,l>, in vector, element is stock price, and parameter l arranges same step 1), is one group, is divided into l/n group with wherein continuous print n knockdown price; Each group calculating mean value, in chronological order, determines r by l/n value point _ivalue, point 3 kinds of situations:

Situation 1. is worth point and continues to rise, or is only no more than 10% drop point of parameter l, and drop range is no more than 2%, now r _i=1;

Situation 2. is worth some continuous decrease, or is only no more than 10% rising point of parameter l, and amount of increase is no more than 2%, now r _i=-1;

Situation 3. does not meet other all scenarios of situation 1 or 2, now r _i=0.

4. the stock certificate data analytical approach based on price linkage network according to claim 3, is characterized in that, above-mentioned steps 2) in according in double bounce father node collection calculate appreciation in share value expect treatment scheme be: given stock node s _a, first consider s _afather node collection wherein each node has sensing s _acontinuous action relation, i.e. directed edge; To set in each node s _i, calculate s _ito s _aappreciation affect e _{<i, a>}, equal directed edge <s _i, s _athe weight w of > _{<i, a>}; Then s is considered _adouble bounce father node collection node wherein needs just can be connected to s through two directed edges _a; To set in each node s _j, calculate s _jto s _aappreciation affect e _{<j, a>}, be s _jto s _aall double bounce paths on, comprise 2 continuous action relations, the maximal value of limit weight product; On this basis, stock s is calculated _aappreciation expect δ _a, formula is as follows:

${\mathrm{\&delta;}}_a=\underset{s_i\&Element;{\mathrm{\&Gamma;}}_a^1\&cup;{\mathrm{\&Gamma;}}_a^2}{\&Sigma;}{r}_i\&CenterDot;e_{<i,a>}$

Wherein r _ifor stock s _itendency mark.