Patent Grant
7542881
The invention relates to the technical field of “Industrial Risk Management” and more specifically to:
The invention relates specifically to the following cases:
These state variables are actually:
Subsequently, and notably in the description and the claims that follow, the term “common state variables” shall be referred to (by extension) as all state variables of either one of the two types described above. Subsequently, the terms random environment and risk environment shall also be used interchangeably.
According to the invention, an industrial site is understood to be any location with human activity where the object is the production of added value (production flow), that is expressed directly or indirectly in terms of added value, whether it be within the manufacturing, energy, food processing, finance, transport, public health, use of raw materials industries amongst others.
Subsequently, as an application example for the invention and an analysis of the previous work, we shall frequently refer to the financial industry. It is within this field that the application of probabilistic prediction processes for random behaviour is more commonly used given that it is governed by regulation constraints (“CAD” directives of the Council of European Communities, “Basle Committee” recommendations, “Group of 30” recommendations). However, lessons taken from this invention may be applied within nearly every industrial field. By using the financial field as an example, we understand quantifiable risk (random environment) to be market, credit, counterparty, liquidity risks amongst others. In contracts, non-quantifiable risks may be legal, fraud or media related risks.
The invention relates specifically to the following cases:
The invention relates to the industrial fields, that are manufacturing and financial, either public or private, where it is suitable:
The second half of the twentieth century became aware of the pernicious risk effects and, as a result, of its cost and the birth of a new science, namely “Risk Management”. “Risk Management” is a science that:
The previous work is concerned with “Risk Management” methods for industrial systems. Thus, U.S. Pat. No. 6,021,402 outlines a risk management method for a network of industrial systems concerned with electric production. U.S. Pat. No. 5,930,762 outlines a risk management system for a device whose behaviour is bound by multiple environment parameters.
The previous work is concerned with five different analysis and/or decision-making methods that relate to risk of random systems:
The present invention relates specifically to the technological field of statistic/probabilistic processes that aim to provide a statistic estimate of the probability law in the distribution of variations for a numerical productivity indicator of that system. The latter is achieved with the help of a probability distribution model together with environment state variables. The more common field of the previous work in which statistic/probabilistic risk analysis processes are used in that of finance, and more notably market risks. Within the global financial world, market risk is one that the investor faces due to his lack of knowledge concerning future evolution of common state variables that are market basics, such as share prices, interest rates, currency rates and so on. These market variables, also known as risk factors, determine prices of shares, quota assets and portfolios amongst others.
Risks associated with different financial products may be assessed through certain characteristic numbers, such as time and convexity of investment bonds, beta and expected performance of shares, time value, volatility, delta, gamma, theta, vega (derived in relation to volatility) and so on for bonds. However, these different risk approaches known as “sensitive”:
It is preferable to have a unified probabilistic and temporal approach to risk. Thus, the previous work used the method known as “Value at Risk” (VaR) as risk indicator for a financial system subjected to a random market environment. For a financial asset (or portfolio), the VaR is an amount such that the loss incurred on that asset (or portfolio) during a certain time interval does not but exceed the VaR with a given (weak) probability level. Calculation of VaR requires knowledge of the probability law for the loss incurred (numerical productivity indicator). The basic problem with risk analysis statistic/probabilistic processes (and especially VaR calculation processes) is therefore to predict distribution of the probability law and distribution of variations for a numerical productivity indicator (loss in the case of VaR).
The statistic/probabilistic processes that aim to provide a statistic estimate of the probability law in the distribution of variations for a numerical productivity indicator of a system are mainly of two types:
Thus, the previous work is concerned with four classical VaR estimate methods (and more generally of statistic estimate of the probability law in the statistic distribution of variations for a numerical productivity indicator of a system subjected to a random environment):
The invention relates solely to the technical fields of “simulation statistic/probabilistic” processes.
In order to calculate the VaR of a financial portfolio, the following entries are classically used:
The “delta normal” (or linear) method, said of the estimated variance-convariance matrix, was suggested and developed by JP Morgan through its RiskMetrix™ system. It consists of three stages as follows:
This method provides the following specific theoretical hypotheses by way of example:
VaR is calculated using the M matrix known as “variance-convariance”, of n×n dimension in the case of n risk factors and consisting of n×(n+1)/2 variances-convariance of risk factor variations calculated on the basis of a past historical period. In that (λ=λii=1 . . . n) represents portfolio sensitivities over different risk factors (X=Xi(i=1 . . . n)), VaR is calculated using the following formula:
VaR=z√{square root over (′λMλ)}+′λ
where z is the quantile of the reduced normal distribution for the confidence interval value of the retained VaR. In such a way that, according to the variance-convariance matrix method, the risk structure is calculated in an environment of n risk factors by generating n′ set parameters of the variance-convariance matrix that describe variances and convariances of risk factor variations. In this sense, the variance-convariance matrix method is an analytical statistic/probabilistic method. The hypothesis of retained linearity meeds strict hypotheses on portfolio sensitivities to be implemented in that it would not be applicable to portfolios with “gamma” or “convexity” risks.
The “delta gamma” method is a more elaborate form of the variance-convariance matrix method. The linearity hypothesis is replaced with a quadratic hypothesis. The method incorporates a second set of sensitivity factors for analysts and requires that application of sophisticated mathematical calculations. Despite its superior position over the “delta normal” methods, the “delta gamma” method is generally unsuitable, especially to assess portfolio risk that includes options reaching maturity or exotic derivative products such as “knock-out” options.
On the contrary, according to the Monte-Carlo simulation method advocated especially by Bankers Trust with its RaRoc 2020™ system:
In this sense, the Monte-Carlo simulation method is a simulation statistic/probabilistic method.
The historical analysis method advocated especially by Chase Manhattan with the Charisma™ and Risk$™ systems is a simplification of the Monte-Carlo method. According to this historical analysis method, a vast sampling number of probable future variations of risk factors are directly transmitted from the market history. By way of example, the historical VaR of one portfolio day may be estimated from one hundred scenario samples in the following manner:
In this sense, the historical analysis method is also a simulation statistic/probabilistic method.
The principle behind risk analysis and the use of a “simulation statistic/probabilistic” method is familiar to the previous work and applied to other fields, both financial and industrial. The present invention, which (in combination) entails the use of this method, does not in any way claim responsibility for this method when considered independently. Patent FR2769703 may be referred to, and outlines a process to determine unpredictable analysis of a device using coordinates according to the Monte-Carlo Method. Patent EP0837408 outlines a pulverisation simulation process using the Monte-Carlo Method. Patent WO9905685 outlines an automatic system for industrial surveillance using the Monte-Carlo Method. Patent WO9859322 outlines a system that generates pixel values of an image using the Monte-Carlo Method. Patent WO9852132 outlines a system for allocating organisational resources and assessing resource costs and prices using the Monte-Carlo Method. Patent WO9850835 outlines a monitoring system that can be programmed to supervise industrial equipment activity using the Monte-Carlo Method. Patent WO9836364 outlines an integral assessment process using the Monte-Carlo Method. U.S. Pat. No. 5,940,810 outlines a technique to estimate complex financial asset value using the Monte-Carlo Method.
The statistic/probabilistic methods that provide an estimate of the probability law in the statistic distribution of variations for a numerical productivity indicator of a system subjected to a random environment (and especially VaR estimate methods) can be divided into two distinct groups:
The invention relates solely to behaviour analysis processes, notably of a “simulation statistic/probabilistic” type that carries out a “selective sampling (selection) of probable scenarios with a multi-value of environment variables”.
The previous work knows how to implement a “simulation statistic/probabilistic” method on a unique site and to notably perform all data acquisitions, statistical calculations and so on in order to determine the probability law in the statistic distribution of variations for a numerical productivity indicator of a system when subjected to a random environment. The previous work also knows how to calculate the VaR of a single financial institution's portfolio by using the “simulation statistic/probabilistic” method.
It is further acknowledged that the previous work is able to repetitively transmit numerical values from a centralised site to a vast number of remote sites. This is especially the case of financial data screens used by Reuters and Bloomberg. A central site is connected to sources of variable and random environment data (share prices, financial assets and so on). This concentrated data is transmitted repetitively in a decentralised way towards a vast number of remote sites (screens of financial operators). However, it will be noted that according to the previous work, the transmitted data is real observable or quasi-observable data (share values), i.e. recreated through calculation (index values, internal productivity rates, and so on), and is to be used as such. The previous work is unaware of the principle of systematic and repetitive remote transmission of probable samplings with a multi-value of possible future values of environment variables with the aim of recreating a histogram.
The methods of the previous work as outlined above do not allow a satisfactory probabilistic prediction of future behaviour and/or the optimisation of productivity parameters for a number of remote local industrial systems subjected to a common global environment to be performed. In particular, they do not allow good VaR conditions of a number of financial sites managing share portfolios to be simultaneously calculated. Thus:
On the one hand, as has been described above, only statistic/probabilistic processes that aim to provide a statistic estimate of the probability law in the statistic distribution of variations for a numerical behaviour indicator deliver accurate and reliable risk assessment.
On the other hand, amongst the statistic/probabilistic processes:
It is acknowledged that the previous work is capable of transmitting the most probable scenario of environment variables towards remote sites (for e.g, televisions). This is especially the case for predictions of meteorological variables (such as winds, temperatures and so on) at various points. However, this does not allow for a global probabilistic view (probability law in the statistic distribution of variations for a numerical behaviour indicator) and is limited to a small number of parameters for each site.
Further, risk management procedures require uniform risk analysis and therefore a unified method in their assessment.
In such a way that the analysis of the previous work emphasises the need for a probabilistic prediction technique of future behaviour (and/or optimisation of productivity parameters) of a vast number of remote local industrial systems (in particular a number of financial sites) subjected to a common global environment. The system should provide an accurate and reliable risk assessment of a “simulation statistic/probabilistic” type, i.e. working with a selective sampling of probable scenarios of environment variables by using a minimum of:
According to the invention, the aim of the “statistic/probabilistic” process by stochastic simulation is to perform in a prioritised way, yet jointly (i.e. simultaneously), the probabilistic prediction of future behaviour and/or the optimisation of productivity parameters of a vast number of remote local industrial systems (Su(u=1 . . . n)). It is applied to the following cases:
According to the invention, the stochastic simulation process follows a classical method of a “simulation statistic/probabilistic” type which initially performs a “selective sampling of probable scenarios of environment variables”. It then performs a statistical estimate of the probability law in the distribution of variations of a numerical productivity indicator for each local industrial system (Su) by using this description sampling of joint probability distribution with state variables in a predetermined environment.
However, according to the invention, the additional aim of the process is to:
According to the invention, and in its more general form, the stochastic simulation process is remarkable in that it combines the following two means:
If necessary, the vector level of the local action parameters (λu)=(λu1, . . . ,λuq) for each local system (Su) is adjusted to an optimal value that is calculated according to predetermined criteria in order to anticipate, in particular, the probable future behaviour and effectiveness of each system (Su) subjected to the environment (E) during any one retained time forecast(s) (tk).
Of the following drawings:
Even local random phenomena, as are climatic phenomena, may not be considered independently from phenomena common to all factories, such as petrol price or the demand level on the network. Thus, temperature close to a factory not only acts upon factory productivity (and therefore its marginal cost) but also interacts with demand level and electricity price. There are strong chances of seeing the local temperature change in correlation with the temperature close to other factories, and therefore to see the latter's marginal cost vary in the same direction and further interact with demand level. Seeking to optimise a factory margin with a statistical analysis of local weather parameters separate from that of global parameters would lead to obvious optimisation mistakes since local and global parameters interact within the same framework of a common global environment.
A series of r captors (γh(h=1 . . . r)) installed at different points in the environment (E) are designed to observe the state of the environment (E). The data provider (Fo) remotely collects at each instant (t) the r measures of the environment status (μh(t)(h=1 . . . r)) from the captors. At each selected sample instant (tk), these r measures constitute temporal vector components to measure the environment (M(tk))=(μh(tk)(h=1 . . . r)). The data provider (Fo) is equipped with a transmitting device (Ao) of observable data. The latter transmits the emission of observable data on to a network (Ro):
Subsequently, observable and quasi-observable variables which we shall globally call “observable variables” are assimilated. The case depicted in
According to the invention, a centralised transmitting site (U) is generally described in
This centralised transmitting site (U) is a great distance (several kilometers, if not several hundred kilometers) from a majority of the n remote local industrial systems (Su(u=1 . . . n)), the power stations (S1, . . . ,Su . . . , Sn), that are distributed over the entire state territory and will appear in
The statistical law (ƒa) with parameters (numerically described) is generally chosen by a mathematician specialising in statistical models. It is determined by historical observations. In particular, it may entail normal laws, normal logs and so on, or more generally, a joint statistic distribution of p environment variables (Xi(tk)(i=1 . . . p)) or their temporal variations (ΔXi(tk)i=1 . . . p)) and is assessed through appropriate and classical statistic tests (averages, variances, correlations, breaks, extreme behaviours and so forth).
As the observable environment data is received, be it by request or continuously, the memory bank (H) is updated in which the periodic history of value measures (μh(t)(h=1 . . . r)) is registered. The calculator (C) calculates the p variables (Xi(t)(i=1 . . . p))=(Φ(μ1(t), . . . , μr(t))) at each successive sampling period with the aid of a descriptive numerical model of the environment (Φ) and the measures vector (M(t)),. Preferably, the calculator (C) calculates and repetitively updates the variations history of p state variables (ΔXi(t)(i=1 . . . p)), which it registers in its memory bank (H). Variations of state variables are understood to be either the difference or ratio of two values of a variable at successive instants, or more generally, any numerical value describing variations (absolute or relative) of the variable between two successive instants.
The calculator (C) is equipped with a simulation system (9) or scenarios, a calibrating system (10) for the statistical law (ƒa) with parameters so that it conforms with stored historical observations in the memory bank (H), and a random generator (11) enabling “smp probabilistic simulations” (Wijk) with multi-value descriptions of probable future pseudo-states or pseudo-variations of state variables (Xi(t)(i=1 . . . p)) to be repetitively created.
Various functioning methods of the scenario simulation system (9) are described further on. Some generate series of simulations of state variables (Xi(t)(i=1 . . . p)) and other, series of simulations of variations of p state variables (ΔXi(t)(i=1 . . . p)). In such a way that, at the centralised transmitting site (U), an “smp series of probabilistic simulation” (Wijk
According to an initial application mode, this “smp series of probabilistic simulation” is an “smp probabilistic series” (Xijk) of pseudo-states formed by s×m p-uplets with a multi-value sampling of future pseudo-states (Xij(tk)(i=1 . . . p,j=1 . . . s,k=1 . . . m)) including possible s×m×p numerical values of probable future pseudo-values for p state variables (Xi(t)(i=1 . . . p)) and constitutes:
According to a second application mode described in
As it appears in
As observed in
According to the invention,
Each local industrial system (Su(u=1 . . . n)) consists of a fairly similar local computer system (13, 14, 15). The make-up of the local computer system (15) for the industrial system (Su) is described in
Note that each local industrial system is equipped with a local satellite dish (16, 17, 18) to receive probabilistic data and a local reception device for probabilistic data. Independently, and within each local industrial system, a succession of “smp probabilistic series” (Wijk) with a multi-value description of probable future pseudo-states or pseudo-variations including s×m×p numerical values (Wijk=Xijk or ΔXijk) are transmitted by the satellite (6) and repetitively collected with the aid of a satellite dish (18) and the local reception device.
Through the local interface (Ωu), m′ (m′≦m) time forecasts are retained from the m said time forecasts (tk(k=1 . . . m)). If necessary (in case 2 described above and represented in
For each retained m′ time forecast (tk), a local systems analysis model (Du) that has been previously set in order to determine the productivity indicator (Iujk=Du(λu,(Xij(tk)) within each s scenario of received simulations (j=1, . . . ,s) is numerically applied:
Thus, the corresponding simulated s pseudo-values of the productivity indicator (Iuj(tk,λu))=Du(λu,(Xij(tk)(i=1 . . . p,j=1 . . . s)) are determined. The corresponding s values of the productivity indicator (Iuj(tk,λu)(j=1 . . . s)) are grouped together and arranged according to the productivity level (Iuj). The histogram (Γ(Iu)) is calculated and stored in the local memory bank (Hu) as is the probability distribution (as a number or percentage) (Ψu(Iu)) of the level reached by the local productivity indicator (Iu) in the configuration (λu) according to the local model (Du). A statistical calculation deduces the value of at least one chosen statistical parameter of adjustment (V(λu)) (such as average, spread type, Value at Risk, utility and so forth) that is characteristic of the probability distribution (Ψ(λu,I)) for productivity levels (Iu(tk,λu) according to the vector of local action parameters (λu)=(λu1, . . . ,λuq).
If necessary, a numerical optimisation (according to predetermined criteria) of one or many chosen statistical parameters of adjustment (V(λu)) is performed within each local system (Su). The vector level of local action parameters (λu)=(λu1, . . . ,λuq) for each local system (Su) is then adjusted to a calculated optimal value according to predetermined criteria, notably in order to anticipate the future probable behaviour and effectiveness of each system (Su) subjected to the environment (E) during any one retained time forecast (tk).
The aim of one particular application mode of the invention described in
The expert site (U′) is linked up to the centralised transmitting site by a network (R′) through which it transmits the adaptive numerical definitions of the productivity components (Iuhkj)=(πh(λuh,Xjk)).
On the one hand, the “smp probabilistic series” (Wijk) with its multi-value description of probable future pseudo-states or pseudo-variations is repetitively transmitted at successive instants (to) from the centralised transmitting site by using the retained joint behaviour statistical model (ƒa(X1, . . . ,Xp;t;Δt)) in time (t) of state variables (Xi(i=1 . . . p)). On the other hand, as each new code (Lh) appears or adjustment to the productivity assessment system (πh) is made, the pair (Lh,πh) is transmitted as a digital file in a centralised way.
At each remote local industrial system (Su), the productivity component (Iuhjk)=(πh(λuh,Xjk)) is calculated:
Subsequently, in summary and according to the above formula (*) of the productivity indicator (Iujk), its histogram is calculated followed by the retained statistical parameter (V(λu)). Finally, the vector of local action parameters (λu) is applied in order to optimise the value of the statistical parameter (V).
A particular case of the above application mode consists in productivity assessment systems (πh) being proportional to the local action parameter (λuh), ie that:
πh(λuh,X)=λuhπh(X)
where (πh) is called the “marginal productivity assessor”. In this case, the local systems analysis model (Du) amounts to creating a scalar product between the vector of local action parameters (λuh) and the vector of marginal productivity (π(X))=(πh(X)(h=1 . . . q)):
Du(λu,X)=(λu|π(X)) (**)
Calculation of the statistical parameter (V(λu)) at the local site (Su) is thereby simplified and its optimisation accelerated by previous registration of the marginal productivity series (πh(Xjk)(h=1 . . . q,j=1 . . . s,k=1 . . . m)) in the memory bank (Hu) of the local calculator (Cu).
A first particular variant of the invention's application is also described in
Xij(tk)=Xi(to)(1+ΔXij(tk))
Any other formula (addition, exponential and so forth) that implicates the “smp variational probabilistic series” (ΔXijk) coming from the centralised transmitting site (U) and the standard value of state variables (Xi(to)(i=1 . . . p)).
A particular application method for this first variant is that described further above with reference to
A second particular combination variant for the application of the invention is described in reference to
The invention recommends two particular methods of repetitive stochastic simulation that prioritises through probable scenario sampling in order to predict probable future behaviour and/or optimise productivity parameters of a number of remote local industrial systems (Su).
The first method suggested by the invention of repetitive stochastic simulation that prioritises through probable scenario sampling consists in its combined use with the method known as Monte-Carlo. According to this first method, which is described in
Said “smp probabilistic series” (Wijk) with a multi-value description of probable future pseudo-states or pseudo-variations is transmitted in a decentralised way from the transmitting site (U) towards all remote local systems (Su). This application method is recommended by the invention.
The second method suggested by the invention of repetitive stochastic simulation that prioritises through probable scenario sampling consists in its combined use with the method known as historical analysis. It is a simplified version of the Monte-Carlo method described above. According to this historical analysis method, the sampling of probable future variations of risk factors is transmitted directly from the memorised history. A history of values for p state variables (Xi(t)(i=1 . . . p)) of the common global environment (E) throughout time is maintained and updated at the centralised transmitting site (U). The “smp probabilistic series” (Wijk) with a multi-value description of probable future pseudo-states or pseudo-variations of environment (E) is numerically generated and repetitively transmitted at successive instants to by using the retained statistical model (ƒa(X1, . . . ,Xp;t;Δtk)) of joint behaviour within the time (t) of state variables (Xi) and the memorised history of values for p state variables (Xi(t)(i=1 . . . p)). According to the method known as “historic”, the estimate of probable future states of p state variables (Xi(t)(i=1 . . . p)) is recreated by using the standard status (Xi(to)(i=1 . . . p)) and past historic variations (ΔXi(t′)(i=1 . . . p)) for a set of past dates (t′<to) and by affecting each recreated state with a production probability that could depend upon the age of the corresponding historic variation. Said “smp probabilistic series” (Wijk) with a multi-value description of probable future pseudo-states or pseudo-variations is transmitted in a decentralised way from the transmitting site (U) towards all remote local systems (Su).
A third application variant of the invention consists in introducing biased data within the transmitted stochastic simulations. The aim of the biased data is take extreme situations into account that only rarely occur and/or constitute “catastrophe scenarios”. According to this method, the generated “smp probabilistic series” (Wijk) with a multi-value description of probable future pseudo-states or pseudo-variations transmitted by the transmitting site (U) towards all remote local systems (Su) is biased in relation to the statistics for the history of state variables or variable variations. According to this variant of the invention, a history of values for p state variables (Xi(t)(i=1 . . . p)) of the common global environment (E) throughout time is maintained and updated at the centralised transmitting site (U). The “smp probabilistic series” (Wijk) with a multi-value description of probable future pseudo-states or pseudo-variations of the environment (E) is numerically generated and repetitively transmitted at successive instants to by using the retained statistical model (ƒa(X1, . . . ,Xp;t;Δtk)) of joint behaviour within the time (t) of state variables (Xi) and the memorised history of values for p state variables (Xi(t)(i=1 . . . p)).
However, joint statistic behaviour biases of state variables (Xi(t)(i=1 . . . p)) in relation to their historic statistic behaviour are deliberately introduced into the “smp probabilistic series” (Wijk). The transmitted biased series thus reproduce “catastrophe scenarios”. This enables the impact of said catastrophe scenarios on remote systems (Su) to be taken into account. Said “smp probabilistic series” (Wijk) with a multi-value biased description of probable future pseudo-states or pseudo-variations is then transmitted in a decentralised way from the transmitting site (U) towards all remote local systems (Su).
According to a first sub-variant of the third variant described above, the invention recommends that the grade of “smp probabilistic series” (Wijk) with a multi-value biased description of probable future pseudo-states or pseudo-variations transmitted by the transmitting site (U) towards all remote local systems (Su) in relation to selective requests made from the different remote systems (Su) be selectively adapted. To this end, certain remote systems (Su) send requests as numerical parameters of catastrophe scenarios according to the specific biases as defined by them to the centralised transmitting site (U). In relation to these requests and by possibly merging those that are identical, “smp probabilistic series” (Wijk) with multi-value descriptions of probable future pseudo-states or pseudo-variations that are biased in relation to specified parameters requested by remote systems are generated at the centralised transmitting site (U). The transmitting site (U) selectively sends each remote local system (Su) said “smp probabilistic series” (Wijk) with a multi-value biased description of probable future pseudo-states or pseudo-variations that corresponds to its request. This is sent electronically (via Internet, local network, telephone or any other method that allows the transmission of digital data).
According to a second sub-variant of the third variant described above, the invention recommends that remote systems (Su) alter all or part of the “smp probabilistic series” (Wijk) they receive from the transmitting site (U) to take possible extreme situations that affect them into account. According to this sub-variant, certain remote systems (Su) alter all or part of the “smp probabilistic series” (Wijk) with a multi-value description of probable future pseudo-states or pseudo-variations. This enables biased series to be created locally by reproducing catastrophe scenarios whose parameters are either specified by the main transmitting site (U) and transmitted to local systems (Su) or defined by the local systems (Su).
A fourth preferred application variant of the invention consists in numerically transmitting the “smp probabilistic series” (Wijk) with a multi-value description of probable future pseudo-states or pseudo-variations from the transmitting site (U) periodically and according to a sampling frequency (φ).
According to a fifth application variant, the invention recommends three methods to transmit the “smp probabilistic series” (Wijk) with a multi-value description of probable future pseudo-states or pseudo-variations from the transmitting site (U) towards all remote local industrial systems (Su(u=1 . . . n)).
A first sub-variant for the emission of probabilistic data is described in
A second sub-variant of the emission of probabilistic data is described in
A third sub-variant of the emission of probabilistic data is described in
According to a sixth application variant, the invention recommends three methods to organise “smp probabilistic series” (Wijk) transmitted from the centralised transmitting site (U) towards all remote local industrial systems (Su(u=1 . . . n)) in order to compress transmitted probabilistic data and to accelerate access by all remote local industrial systems (Su(u=1 . . . n)).
According to a first sub-variant for organising the emission of probabilistic data, an “smp probabilistic series” (Wijk) with its multi-value description of probable future pseudo-states or pseudo-variations of the environment (E) is numerically generated and repetitively transmitted at successive instants (to) from the centralised transmitting site using the retained joint behaviour statistical model (ƒa(X1, . . . ,Xp;t;Δt)) in time (t) of state variables (Xi(i=1 . . . p)). This series consists of s (where s is much greater than >> 1) numerical pseudo-values for p components (Wij(tk)(i=1 . . . p,j=1 . . . s)) for each m (m≧1) time forecast (tk(k=1 . . . m)) as an “smp file” of s×m×p successive alpha-numerical data. However, on the one hand, this “smp file” is sub-divided into successive m “temporal sub-files” of level 1 that groups together all pseudo-values of a same time forecast (tk(k=1 . . . m)). On the other hand, each “temporal sub-file” consists of a succession of s “selection sub-files” of level 2 each made up of a pseudo-value p-uplet.
According to a second sub-variant for organising the emission of probabilistic data, an “smp probabilistic series” (Wijk) with its multi-value description of probable future pseudo-states or pseudo-variations of the environment (E) is numerically and repetitively generated at successive instants (to) from the centralised transmitting site (U) by separating successive m “temporal sub-files” of level 1 and gathering together all pseudo-values of a same time forecast (tk(k=1 . . . m)) with a corresponding time forecast indicator (ηkk=1 . . . m)). At each local system (Su), different said temporal sub-files of a same “smp probabilistic series” (Wijk) with its multi-value and collected data are separated from one another by identifying their specific time forecast indicators (ηkk=1 . . . m)).
According to a third sub-variant for organising the emission of probabilistic data, an “smp probabilistic series” (Wijk) with its multi-value description of probable future pseudo-states or pseudo-variations of the environment (E) is numerically and repetitively generated at successive instance (to) from the centralised transmitting site (U) by separating successive s “selection sub-files” of level 2 (from a same temporal file) and each gathering together a pseudo-value p-uplet with a corresponding selection indicator (τj(j=1 . . . s)). At each local system (Su), different successive “selection sub-files” of level 2 (from a same temporal file) each gathering a pseudo-value p-uplet from a same “smp probabilistic series” (Wijk) with its multi-value and collected data are separated ones from the other by identifying their specific selection indicator (τj(j=1 . . . s)).
According to a seventh application variant, the invention recommends three method to selectively filter the content of the “smp probabilistic series” (Wijk) transmitted by the centralised transmitting site (U) towards all remote local industrial systems (Su(u=1 . . . n)). It aims to adapt their content to the needs and features of the remote local industrial systems (Su(u=1 . . . n)) as well as to the hierarchical, industrial or commercial ties between the centralised transmitting site (U) and all remote local industrial systems (Su(u=1 . . . n)).
A first sub-variant for filtering the emission of probabilistic data is described in
A second sub-variant for filtering the emission of probabilistic data is described in
A third sub-variant for filtering the emission of probabilistic data is described in
According to an eighth application variant described in
At each n′ of retained local systems (Su) and for each m′ of retained time forecast (tk), the local systems analysis model (Du) is numerically applied having been previously set in order to determine the productivity indicator level (Iujk=Du(λu,(Xi,j(tk))) within each s scenario (j=1, . . . ,s):
The simulated s pseudo-values corresponding to the productivity indicator (Iuj(tk,λu))=(Du(λu,Xj(tk))(j=1 . . . s)) are determined. The “local simulation series” consisting of s simulated pseudo-values corresponding to the productivity indicator (Iuj(tk,λu))=(Du(λu,Xj(tk))(j=1 . . . s)) of the local system (Su) is transmitted in a centralising way towards a centralising site (B).
At the centralising site (B), the simulating s×n′ pseudo-values of the productivity indicators (Iu(tk,λu))(j=1 . . . s)) are collected from each n′ local system (Su) and for each s simulation, and the global productivity indicator (Jj) of all local systems is calculated for each simulation by using a global assessment model (Jj)=(χ(Iuj(tk,λu)(u=1 . . . n′,j=1 . . . s)). The corresponding s values of the productivity indicator (Jj(tk;λ1, . . . ,λn′)(j=1 . . . s)) are grouped together and arranged according to their level. The histogram (Γ) is calculated and stored as is the probability (as a number or percentage) distribution (Ψ(J)) of the level reached by the global productivity indicator (J) in the configuration of local action parameters vector (Λ)=(λu(u=1 . . . n′)) chosen according to the retained local models (Du) and the global assessment model (χ). A statistical calculation deduces the value of at least one chosen statistical parameter of adjustment (V(Λ)) (such as average, spread type, Value at Risk, utility and so forth) that is characteristic of the probability distribution (χ(λ1, . . . ,λn′,J)) for global productivity levels (Jj(tk;λ1, . . . ,λn′)) in relation to the multi-vector values (□)=(□1, . . . ,□n) of the vectors of local action parameters (□u).
If necessary, a global numerical optimisation (according to predetermined criteria) of the chosen statistical parameters of adjustment (V(□)) is performed in relation to the multi-vector (□)=(□1, . . . ,□n). The corresponding level of its vector optimised from the local action parameters (λu) is transmitted to each local system (Su) in order to globally anticipate the future probable behaviour and effectiveness of all local system (Su) subjected to the environment (E) during any one retained time forecast (tk).
According to the invention,
For example, the value (Iu(u=1 . . . n)) of a portfolio (Pu(u=1 . . . n)) may be calculated according to the following linear combination:
where λh(t,X(t),θh) is the asset price (θh) art instant (t) according to the local model (Du) when the market is in state (X(t)), or by using any other formula that allows for a non-linearity in relation to quantities (λuh) particularly in order to take liquidity risk into account.
According to a first sub-variant of the invention's process of application to the financial industry, the following prior choices are made at the expert site (U′):
The statistic model of joint behaviour in time (t) and by time periods (Δtkk=1 . . . m)) of market variables is determined from historical observations that may include, in particular, normal laws, normal logs and so on, or more generally, a joint statistic distribution of p environment variables (Xit(k)(i=1 . . . p)) or their temporal variations (ΔXi(tk)i=1 . . . p)). It is assessed through appropriate and classical statistic tests (averages, variances, correlations, breaks, extreme behaviours and so forth).
At the centralised transmitting site (U), and by using the statistical model (ƒa(X1, . . . ,Xp;t;Δtk)) retained from joint distribution within the time (t) of market variables (Xi(i=1 . . . p)), an “smp probabilistic series” (Wijk) with a multi-value description of pseudo-states or pseudo-variations for the probable market future (E) is numerically generated at repetitive and successive instants (to).
According to a first application mode, this “smp probabilistic series” (Wijk) is an “smp probabilistic series” (Xijk) of pseudo-states formed by s×m p-uplets of a multi-value sampling for future pseudo-states (Xij(tk)(i=1 . . . p,j=1 . . . s,k=1 . . . m)) that include possible s×m×p numerical values of p market variables (Xi(t)(i=1 . . . p)) and constitute:
According to a second application mode, this “smp probabilistic series” (Wijk) is an “smp variational probabilistic series” (ΔXijk) formed by s×m p-uplets of a multi-value sampling for future pseudo-variations (ΔXij(tk)(i=1 . . . p,j=1 . . . s,k=1 . . . m)) from the standard state of market variables (Xi(to)(i=1 . . . p)) that include s×m×p numerical values of probable future pseudo-variations for p state variables (Xi(tk)(i=1 . . . p)) and constitutes:
Thus, the process of the invention, at the centralised transmitting site (U) that is quite a distance from the majority of n local financial operators (Su(u=1 . . . n)), is to transmit at repetitive and successive instants (to) said “smp probabilistic series” (Wijk) with a multi-value description of probable future pseudo-states or pseudo-variations that includes s×m×p numerical pseudo-values and consists of:
This transmission of said “smp probabilistic series” (Wijk) is performed in a decentralised way from the transmitting site (U) towards all remote local operators (Su).
Independently, and from each of the local operators (Su), the majority of which are quite a distance from the centralised transmitting site (U), m′ (m′≦m) time forecasts retained from amongst the m said time forecasts (tk(k=1 . . . m)) are chosen. If necessary, as in case 2 in which only said “smp variational probabilistic series” (ΔXijk) of variational sampling is received, said corresponding “smp probabilistic series (Xijk) of pseudo-states” consisting of pseudo-samplings of future states (Xij(tk)(i=1 . . . p,j=1 . . . s,k=1 . . . m)) (at least for the retained m′ time forecasts (tk)) is locally calculated. Subsequently, for each retained m′ time forecasts (tk), the global value (Iujk=Du(λu,(Xij(tk)) of the portfolio (Pu) is numerically calculated according to a local assessment model (Du) that has been previously set within each s scenarios (j=1, . . . ,s). The calculation is done for at least one particular value of the quantities vector (λu)=(λu1, . . . ,λuq) of held assets and for each simulated s p-uplets of state variables (Xij(tk)) that corresponds to said retained time forecast (tk) for the probabilistic pseudo-state series (Xij(tk)(j=1 . . . s)). The corresponding simulated s pseudo-values of the portfolio (Iuj(tk,λu))=Du(λu,(Xij(tk)(i=1 . . . p,j=1 . . . s)) are then determined. These s portfolio (Pu) values are grouped together and arranged. The histogram (Γu) is then calculated and stored as is the probability (as a number or percentage) distribution (Ψu(Iu)) of the local portfolio (Pu) value (Iu) in the configuration (λu) according to the local model (Du). A statistical calculation deduces the value of at least one chosen statistical parameter of adjustment (V(λu)) (such as average, spread type, Value at Risk, utility and so forth) that is characteristic of the probability distribution (Ψ(λu,I)) for values (Iu(tk,λu) in relation to vector variations (λu)=(λu1, . . . ,λuq) describing the portfolio (Pu).
If necessary, a numerical optimisation (according to predetermined criteria) of one or many chosen statistical parameters of adjustment (V(λu)) is performed locally within each local operator (Su). The portfolio composition (λu)=(λu1, . . . ,λuq) of each local operator (Su) is then adjusted to a calculated optimal value in relation to predetermined criteria notably in order to anticipate the future probable behaviour and yield of each portfolio (Pu) subjected to the market (E) during any one retained time forecast (tk).
According to a second sub-variant of the invention's process of application to the financial industry, the value (Iujk) of a financial operator's (Su) portfolio (Pu) at a future date (tk) according to the model (Du) and within the state of the simulated market (Xj(tk)) that corresponds to scenario j is assessed by taking the simulated market states in the same scenario j at subsequent dates (tl(l=k+1 . . . m)) into account in the model (Du) according to the following principles:
This particular arrangement allows the liquidity risk to be taken into account by imposing upper bound constraints to the quantity of each asset (θh) that can be rightly assessed according to the local model (Du) at any given date (tk), and thus leading to a possible report of the resulting virtual position within each simulation over subsequent dates.
The invention has an industrial application in all fields, such as manufacturing, energy, food, finance, transport, public health, use of raw materials and any other industries that are subjected to a common global environment and where it is suitable to measure and/or control the impact of possible changes in global and local environment variables on the production of a number of remote industrial sites.
By way of example, the lessons of the invention may be favourably used to optimise the watering of crops, calculate road travel time or manage a fleet of cars.
Thus, a first application of the invention relates to agriculture. For this application, the remote local industrial systems (Su(u=1 . . . n)) are the farms of different plots of cultivated land. State variables (Xi(t)(i=1 . . . p)) of the environment are air temperature, pluviometry, hygrometry, dryness of land and others at different geographical points. The local action parameters (λu)=(λu1, . . . ,λuq) are mainly the characteristics of plots and plantations. The aim of the invention is to limit over-consumption of water by automatically activating different watering systems only when the probability of arable lands suffering from dryness within the coming hours or days is above the acceptance threshold as defined by the farmer of each plot of land. Thus, the automatic watering systems are not activated if the probability of rain in the coming hours is high. Conversely, they could be activated one evening, for example, if the probability of extreme dryness the following day is high. Mutual cooperation between those that generate the sampling of meteorological variables and therefore, the expenses of the centralised probabilistic transmitting site (U) means that the setting-up of this automatic watering activation system is less costly.
A second application of the invention relates to optimising road travel. The target is to estimate the probability law on time taken for a driver or a haulage contractor to reach their destination given all external random factors. For a driver needing to reach a point at a specific time, the optimisation of complex probabilistic criteria (probability of reaching a set point at a given time) is the only aspect which will allow them to truly optimise their chosen itinerary. Statistically, there could indeed be a section of the road where traffic generally moves freely, yet when it slow downs, the road may be jammed for a long time, and the view of the average (expected) travel time is therefore too simplistic. According to the invention, each vehicle of a road network consists of a remote local industrial system (Su). The vehicles are subjected to random factors (Xi(t)(i=1 . . . p)) which include, for example, traffic fluidity on different sections of the road, the meteorologic situation, an unexpected breakdown or accident and so forth. The local action parameters (λu)=(λu1, . . . ,λuq) of each vehicle represent the situation of the road network, planned route, speed of vehicle, type of vehicle and driving, and so forth. At the different remote industrial systems, the interface (Ω) for capturing parameters (λu1, . . . ,λuq) may be linked up to different control systems in the vehicle (notably a satellite navigation system, such as GPS, and vehicle captors) and a probabilistic data receiver (Rpu) may be connected to a loaded Internet terminal.
A variation to this second application relates to managing a fleet of vehicles (public transport, freight transport, car hire amongst others). If each vehicle or rolling system is fitted with a localisation system (of GPS type for example) and a transmitter, then it is possible to know the distribution of the fleet at a past or present instant. It is also possible to know “the most probable situation” (expectation) of vehicle location at any given future date. By centralising the simulated pseudo-values of the productivity indicators (Iuj(tk,λu)) at a centralising site (E), the invention allows the manager of vehicle numbers to hone his vision on the probability of coming events. Thus, the manager of vehicle numbers may automatically control the probability of unexpected events that are rare yet laden with consequences (probability that all trucks of a freight haulage contractor simultaneously break down or probability that all vehicles of a car hire service are concentrated in one area).
A third particularly favourable application of the invention relates to the management of parameters for industrial units working on a same network and subjected to a common global random environment. By way of example, the lessons of the invention may be favourably applied to the field of electricity production within a deregulated environment where the price of electricity is set by comparing offer of n producers (Su(u=1 . . . n)) and demand by a high number of consumers on a central market (the “energy pool”). The meeting of offer and demand may be done on the basis of a main planning centre that provides each producer with its working parameters. Such an approach poses the following two problems:
This approach, which is predominantly used in the field of electricity production, was analysed by the applicants. This study found that, in practice, the approach led to significant industrial and energy overcapacities.
The classical alternative to relocate the optimisation problem is to establish a market where offer and demand can meet with a balanced price. The economic theory shows that searching for local optimal values relates to a global optimal value under certain very general hypotheses of convexity. The problem is technically laborious for electricity since storage is impossible. Thus, there are just as many markets as there periods during the year. The recent general move towards deregulating electricity markets has currently led to several market organisation models that it would be too long-winded to list them here. However, they more or less meet along the same lines as the model described below.
The “energy pool” is segmented into (m) sub-markets that correspond to separate time forecasts (Δtk(k=1 . . . m)) with a “notice period”. By way of example, we distinguish a very short-term market of a few minutes intended to maintain tension on the network, a short-term market of 1 hour, namely a “balancing market”, intended to compensate for offer and demand uncertainties, a “normal” market of 24 hours and so on. For every sub-market, each producer (Su(u=1 . . . n)) (power station) must be committed to respect a production cost curb (Cku(w)) that provides production costs for cost capacity level (w) offered on the network (this cost may be infinite beyond a maximum capacity level). For every sub-market, the “energy pool” classifies the capacity offers according to increasing cost and matches them up with demand on the particular sub-market. where the price (Pk(k=1 . . . m)) of a kilowatt/hour on the sub-market is a result of this match. Producers with production costs above the price (Pk) are not called upon by the network manger. If a producer, following accidental unavailability, notices during the time (Δtk) before delivery that it is unable to respect its commitment on the sub-market (l) (with Δtk<Δtl) then it must buy back the corresponding quantities on the sub-market (k) (or any other sub-market wth a shorter notice period). within this application framework of the invention, the producer's productivity indicator is its gross margin. A producer's (Su(u=1 . . . n)) order parameters (λu=(λu1, . . . ,λuq)) are specifically the following:
The choice of parameters is subjected to local operating constraints such as time and minimum start-up costs for a unit, change of fuel type, minima and maxima techniques, problems of discharge limits that pollute air and water, and so forth. The local optimisation problem is complex, insofar as its calls upon a forecast based on random variables that are relevant to both an environment specific to the installation (availability of production installations) and a global environment common to all producers, including the following:
This global environment is very familiar to all producers where prices are the same for everyone and climatic variables not only strongly correlate between different sites but also interact with price booms (for example, we expect a very cold day predicted within 24 hours would reflect a surge in electricity prices on the market within 24 hours).
The global optimisation problem by a main planning system is replaced by n local optimisation problems, in order to maximise the expected gross margin for example and to guarantee a minimal gross margin of x% to bear the debt charge (typically, a Value-at-Risk calculation), thereby entailing complex calculations. According to the previous work, the three possible solutions are as follows:
The application of the invention's lessons allows each electricity producer to choose the third approach, i.e. the only one that allows an exact search for a local optimal value all the while mutualising the complex side of calculations. According to the invention:
The complex optimisation problem in a random environment is reduced to a simple optimisation problem that is to seek the maximum of Embu(λu) under the constraint VaRu(λu)<Vmax for each power station (Su(u=1 . . . n)).
The invention allows for computing systems and human resources necessary for the application of probabilistic simulations on a set of remote industrial sites that are subjected to a common random environment to be minimised with the notable end results being:
Another advantage of the invention is to simplify the global risk calculation of a group of remote local industrial systems.
A third advantage of the invention is to perform local statistical simulation calculations at a number of industrial sites by using mutualised random selections without any local system having to transmit information to the centralised transmitting site on its situation or having to adjust its industrial parameters. This ensures an improved industrial and commercial confidentiality.
In addition, the application example in the electric energy field given above allows the following advantages for an industrial producer to be outlined:
The applicants reserve themselves the sole right to the invention, its description and relevance now justified through detailed examples, throughout the duration of the patent, with no limitation other than that contained in the terms of the claims hereinafter.
| Number | Date | Country | Kind |
|---|---|---|---|
| 00 06113 | May 2000 | FR | national |
| Filing Document | Filing Date | Country | Kind | 371c Date |
|---|---|---|---|---|
| PCT/FR00/02055 | 7/17/2000 | WO | 00 | 1/28/2003 |
| Publishing Document | Publishing Date | Country | Kind |
|---|---|---|---|
| WO01/86473 | 11/15/2001 | WO | A |
| Number | Name | Date | Kind |
|---|---|---|---|
| 5826065 | Hinsberg et al. | Oct 1998 | A |
| 6061662 | Makivic | May 2000 | A |
| 6085175 | Gugel et al. | Jul 2000 | A |
| 6125355 | Bekaert et al. | Sep 2000 | A |
