DE102010046067A1 - Computer-implemented method for automatically adapting stimulated marketplace data at actual marketplace data, involves interconnecting input of block with output of model while maintaining automatically specified value consequence - Google Patents

Abstract

The method involves providing a computation block with a computation prescription, and computing simulated marketplace data by the block. The computation prescription is automatically determined depending upon a computation parameter that is determined on a value consequence, so that given mistake criterion is minimum with reference to actual marketplace data (Sj) and adapted the simulated marketplace data. An input of the computation block is interconnected with an output of a purchaser model while maintaining the automatically specified value consequence. Independent claims are also included for the following: (1) a computer program product for implementing and executing a method for automatically adapting stimulated marketplace data at actual marketplace data (2) a computer for automatically adapting stimulated marketplace data at actual marketplace data.

Description

The invention relates to a computer-implementable method for the automatic adaptation of market data V _j simulated with erroneous survey values to real market data S _j and to a computer program product for implementing and carrying out the method. Moreover, the invention relates to a computer for the automatic adaptation of market data V _j simulated with erroneous survey values to real market data S _j .

Out Gary L. Lilien, Arvind Rangaswamy, Marketing Engineering, 2nd Edition, Trafford Publishing, 2004 Various buyer models are known to simulate market data. Market data are in particular market shares or number of shares. A market share is the percentage of the turnover of a product in the total turnover of the market. A lot number is the percentage of the sold lots of a product in the total pieces of the market.

As a result, real market data S _j are generally percentages with the number n, which have the relative standard deviation δ and which, as a sum, yield 100%. The relative standard deviation δ is the standard deviation of all values S _j normalized to the mean value:

In the same way, also the simulated market data V _{j are} percentages with the number n, which have the relative standard deviation β and which as a sum yield 100%. The relative standard deviation β is the standard deviation of all values V _j normalized to the mean value:

All buyers' models have in common that the buyer models have concrete numerical values as input parameters. The output parameters are the simulated market data. By modifying the input parameters, one can examine how a change in certain products affects the market data.

The input parameters are again based on survey values. These survey values must have a certain accuracy in order to perform calculations with the buyer model. If a certain accuracy of the survey values no longer exists, then the stability and convergence of the entire calculation of the simulated market data is no longer guaranteed.

For this reason, the survey values must be based on a certain minimum sample size. As a rule, it is necessary to include at least 100 respondents in the sample size. This means that the effort for determining the input parameters is very high.

The object of the invention is therefore to increase the accuracy, the stability and the convergence in the calculation of the simulated market data, in particular when there are faulty survey values.

This object is achieved according to the method according to the invention in that a calculation block with a calculation rule T (V _j ) is provided, wherein the calculation block calculates simulated market data P _j = T (V _j ) with the number n from the simulated market data V _j . wherein the calculation rule T (V _j ) depends on a calculation parameter F (V _j ) and the calculation block automatically sets the calculation parameter F (V _j ) to a value sequence F ₀ (V _j ), so that a predetermined error criterion with reference to the real market data S _j and the adjusted simulated market data P _{j is} minimal, and wherein the input of the calculation block is connected to the output of the purchaser model while maintaining the automatically fixed value sequence F ₀ (V _j ).

In the same way, the object is achieved according to the computer according to the invention in that the computer has a calculation block with a calculation rule T (V _j ), wherein the calculation block from the simulated market data V _j adapted simulated market data P _j = T (V _j ) with the Number n and the calculation rule T (V _j ) depends on a calculation parameter F (V _j ) and the calculation block automatically sets the calculation parameter F (V _j ) to a value sequence F ₀ (V _j ), so that a predetermined error criterion with reference to the real market data S _j and the adjusted simulated market data P _{j become} minimal, and wherein the computer has an interface via which the input of the calculation block can be interconnected with the output of the buyer model, while maintaining the automatically determined value sequence F ₀ (V _j ).

The invention is based on the recognition that the buyer model with the faulty survey values is adapted to real market data by a downstream calculation block. The calculation block becomes an integral part of the purchaser model while maintaining the automatically determined value sequence F ₀ (V _j ) by connecting the input of the calculation block to the output of the buyer model. The buyer model thus continues to calculate the simulated market data V _j , but this is now passed on to the calculation block, which then calculates the adapted simulated market data P _j from the simulated market data V _j . By changing the input parameters in a targeted manner, it is still possible to investigate how a change in certain products affects the now adapted simulated market data P _j .

In principle, the method according to the invention can be applied to all types of buyer models in which simulated market data V _{j are} calculated at the output, since the inventive interconnection of the calculation block corresponds to an interconnection of two devices. The existing device initially supplies the simulated market data V _j . After the interconnection with the calculation block, a quasi new device is created which supplies the adapted simulated market data P _j . The old device will continue to perform the operations intended for that device. However, the technical effect of the interconnection means that the results of these operations now have a higher accuracy. The deviation between the real market data S _j and the adjusted market data P _j after interconnection is thus less than the deviation between the real market data S _j and the simulated market data V _j .

Without limiting the generality, the usual percentages are assumed here as market data, ie the real market data S _j are percentages with the number n, which have the relative standard deviation δ and which, as a sum, yield 100%. In the same way, then also the simulated market data V _{j are} percentages with the number n which have the relative standard deviation β and which yield 100% as a sum and the adjusted simulated market data P _j are percentages with the number n which is the relative standard deviation γ and the sum of 100%. Of course, however, any values are conceivable, which are then normalized to percentages by dividing all values by the value sum.

wobei der Berechnungsparameter F(V_j) für alle V_j den gleichen Wert F annimmt.According to a preferred embodiment, it is provided that the calculation block calculates the adjusted simulated market data P _j according to

wherein the calculation parameter F (V _j ) assumes the same value F for all V _j .

The base e is the Euler number. This calculation rule is known from logistic regression ( Gary L. Linien, Arvind Rangaswamy, Marketing Engineering, 2nd Edition, Trafford Publishing, 2004, p. 46 ). However, it is not known with this calculation rule to realize an efficient and stable adaptation method in order to improve the accuracy of existing buyer models by the method according to the invention.

Since the calculation parameters F (V _j) for all _j V takes the same value F, it follows that the automatically set sequence of values F ₀ (V _j) for all _j V have the same value R assumes _0. By using only a single calculation parameter F for all V _j , it is possible that automatic determination can proceed with high efficiency and safety. According to a preferred embodiment, it is provided that the automatic determination of the value F _{0 is carried out} by an iteration routine in which the error criterion is gradually minimized by decreasing changes .DELTA.F with respect to the value F, wherein the iteration routine terminates when the changes .DELTA.F fall below a predetermined resolution accuracy to have. Since the iteration routine only has to optimize one adaptation parameter, the iteration routine is guaranteed to converge even with extremely error-prone poll values.

Surprisingly, it has been shown that the adjusted simulated market data P _j after the adaptation and with a targeted change of the input parameters also deliver usable results if the survey values are related only to the sample size with a single respondent. This means that occasionally surveys can be dispensed with entirely if the survey results are provided by an expert with sufficient market knowledge.

If only one parameter F is used in the adaptation, the parameter F must act on all the simulated market data V _j simultaneously in a suitable manner. This is achieved by varying the relative standard deviation β of the simulated market data V _j by the adaptation parameter F approximately proportionally according to γ ~ F · β. For the relative standard deviation γ of the adapted simulated market data P _{j, the} following applies:

It can be shown that when applying the above calculation rule T (V _j ), the following relationship exists between the relative standard deviation γ of the adjusted simulated market data P _j and the relative standard deviation β of the simulated market data V _j . γ = F / n · β, F ≥ 0.

The proportionality factor is thus F / n, whereby the proportionality factor is only influenced by F, since the number n of the percentage values remains constant.

The above calculation rule T (V _j ) of the calculation block with the automatically determined value F ₀ thus pulls the simulated market data V _j accordion-like together or apart, in such a way that the predetermined error criterion with respect to the real market data S _{j is} minimal. The resulting deviation between γ and δ is thus less than the deviation between β and δ.

If the automatically determined value sequence F ₀ (V _j ) assumes the same value F ₀ for all V _j , then the automatic determination of the value F ₀ can be carried out as described by a simple iteration routine. The disadvantage of a constant value F ₀ , however, is that the deviations (S _{j -Pj} ) can not be reduced to zero as a rule. Compensation of the deviations (S _j -P _j ), however, is possible if according to the invention a variable calculation parameter F (V _j ) is used. As a calculation rule T (V _j ), for example, the calculation rule of the above embodiment can be used analogously by replacing the constant calculation parameter F by the variable calculation parameter F (V _j ):

The variable calculation parameter F (V _j ) can then be automatically set to a value sequence F ₀ (V _j ) containing interpolation points F _j for which within a given resolution the resulting deviation (S _j -P _j ) becomes zero. If specific input parameters of the purchaser model are purposefully changed after the interconnection of the calculation block, then it is also necessary that values outside the interpolation points F _{j are also} defined for the value sequence F ₀ (V _j ). There are different variants here.

A variant consists in that the value sequence F ₀ (V _j ) outside the interpolation points F _j is determined by connecting or interpolating the interpolation points F _j as a function of the value V of the simulated market data.

Another variant consists in that the value sequence F ₀ (V _j ) outside the interpolation points F _j is determined by the fact that the corresponding interpolation point F _{j is} constantly continued for each parameter j and thus assumes the same value F for the corresponding market data value V _j . The parameter j is the index used to denote the relevant market data (ie V _j , S _j , P _j ).

In a further variant, the value sequence F ₀ (V _j ) outside the interpolation points F _j can be determined both as a function of the value V of the simulated market data and as a function of the product parameter j. The value sequence F ₀ (V _j ) is then a function of both the variable V and the index j.

In all the variants described, the calculation of the adjusted simulated market data P _j from the simulated market data V _j proceeds in principle in the same way. For the simulated market data V _j , the respective calculation parameter F (V _j ) from the set value sequence F ₀ (V _j ) is first determined as a function of the value V and of the index j. With these calculation parameters F (V _j ), the calculation rule T (V _j ) in the calculation block is then executed. The result is the adapted simulated market data P _j .

vorgegeben werden. Dieses Fehlerkriterium entspricht der Methode der kleinsten Quadrate.The error criterion can be determined, for example, by the sum of the error squares according to

be specified. This error criterion corresponds to the method of least squares.

Another possibility is that the error criterion is given by a sum of normalized error squares according to

In contrast to the method of least squares of the errors (S _j - P _j) is this error criterion is based on the function f (S _j) based. Thus, by a suitable choice of f (S _j ), it can be achieved, for example, that the relative error and not the absolute error tends to be evaluated.

According to a further preferred embodiment, it is provided that specific input parameters of the buyer model are selectively changed after the interconnection of the calculation block. Increasing the accuracy of the interconnection now makes it possible to simulate certain market situations, even with the error-prone survey values. For example, it is possible to investigate how price changes affect the number of shares or the market shares. Since the unit prices of the products in each buyer model are taken into account as input parameters, it is also possible to change a single unit price or the relevant input parameter in a targeted manner.

With a computer program product, the inventive method can be implemented and executed on a computer.

Further advantages and details of the invention will be described with reference to the following embodiments:

1 shows the functions f ₁ (S _j ), f ₂ (S _j ) and f ₃ (S _j ),

2 shows the basic interconnection of the calculation block with the buyer model,

3 2 shows a profile of the calculation parameter F (V _j ), in which the calculation parameter F (V _j ) assumes different values F for different values V _j ,

4 shows the course of the product F (V _j ) * (V _j ) for the values of the calculation parameter F (V _j ) according to 3 .

5 shows opposite 3 another course of the calculation parameter F (V _j ), in which the calculation parameter F (V _j ) for different values V _{j takes} different values F, and

6 shows for a buyer model with interconnected calculation block the course of the simulated market shares, after the unit prices were changed one after the other purposefully.

Embodiment 1

1 shows the functions f ₁ (S _j ), f ₂ (S _j ) and f ₃ (S _j ) as possible functions f (S _j ) for the error criterion

If for f (S _j ) the function f ₁ (S _j ) = S _j is selected, the error criterion is then based on the relative error exactly. Here, however, it has to be considered that the evaluation for S _j → 0 is very heavily weighted and that the error criterion for S _j = 0 has no solution.

To solve the problem for S _j → 0, the following function f ₂ (S _j ) can be selected:

The base e is again the Euler number. f ₂ (S _j ) is symmetric about the value S _j = 0.5. A favorable value for the parameter a is the value 2. In 1 Therefore, the value a = 2 was chosen for the parameter a of the function f ₂ (S _j ). For a = 2, the slope of f ₂ (S _j ) for S _j = 0.5 assumes the value 1, so that in the middle range the error criterion is again based on the relative error. On the other hand, the function f ₂ (S _j ) flattens out in the outer regions, so that there is also a solution for S _j = 0, for example, f ₂ (0) = 0,1192 for a = 2.

Another possible function for f (S _j ) is finally the following function: f ₃ (S _j ) = 0.5 * (1-b) + b * S _j

This straight line is also symmetric about the value S _j = 0.5, where the parameter b specifies the slope. A favorable value for the slope b is the value b = 0.6. In 1 Therefore, the value b = 0.6 was chosen for the parameter b of the function f ₃ (S _j ).

Embodiment 2

2 shows the basic interconnection of the calculation block with the buyer model. The buyer model in this example is the so-called Constant Decision model, in which for given product features the buyer interest of the product features is linked to the product strengths for each product feature.

The products in this embodiment are intended to relate to drilling machines without limitation to the generality. There are four products, the number of pieces S _j , the unit prices c _j and the market shares Y _{j of} the four products are known: Table 1 Market segment with 4 drills product 1 2 3 4 known number of pieces S _j 0.224 0.361 0.277 0.138 known unit prices c _j in EUR 130.00 170.00 110.00 90.00 known market shares Y _j 0.22 0.46 0.23 0.09

In general, the following relationship exists between a market share Y _j and the number of unit shares S _j and the unit prices c _j :

The four products are to be examined with reference to the three product characteristics 1 = price, 2 = functions and 3 = quality with the Constant Decision model. A marketing expert sets the following input parameters for the Constant Decision model: Table 2 Market segment with 4 drills X position strength W Buyer interest 1 2 3 4 price 0.2 0.1 0.3 0.4 0.3 features 0.2 0.4 0.3 0.1 0.5 quality 0.4 0.3 0.2 0.1 0.2

In matrix form, the values of the buyer interest can be written as follows:

Each value of a buyer's interest corresponds to a poll value in terms of interest in one of the product features 1, 2 or 3. The sum of all values is therefore 1. For example, the marketing expert believes that for 30% of all buyers, price is an important feature.

In matrix form, the position strength values can also be written as follows:

The rows of the matrix X each relate to a product feature. The positional strengths have been estimated by the marketing expert such that the sum of the rows is one each, and that the ranking of the row values corresponds to the ranking of the customer's interest in a product with respect to a product feature.

The multiplication of the matrix W with the matrix X yields:

The column sums of the matrix R yield the simulated market data V _j according to the constant-decision model: V _j = (0.24 0.29 0.28 0.19)

mit f(S_j) = f₃(S_j) und b = 0,6 gemäß Ausführungsbeispiel 1 sowie aufgrund von

berechnet die Iterationsroutine des Berechnungsblocks automatisch den Wert F₀ = 8,77. From the condition

with f (S _j ) = f ₃ (S _j ) and b = 0.6 according to Embodiment 1 as well as due to

the iteration routine of the calculation block automatically calculates the value F ₀ = 8.77.

The iteration routine returns the value 0.01117684 as the minimum value for the error criterion.

The adjusted simulated market data is calculated as:

Tabelle 3 Marktsegment mit 4 Bohrmaschinen Produkt 1 2 3 4 angepasste simulierte Stückzahlenanteile P_j 0,2167 0,3359 0,3077 0,1397 simulierte Marktanteile Y_j, berechnet aus P_j 0,21 0,43 0,26 0,10 From the known unit prices, the corresponding market shares can be calculated according to:

Table 3 Market segment with 4 drills product 1 2 3 4 adapted simulated number of pieces P _j .2167 .3359 .3077 0.1397 simulated market shares Y _j , calculated from P _j 0.21 0.43 0.26 0.10

A comparison with the given values in Table 1 and the simulated values in Table 3 already shows a good match.

Embodiment 3

The deviation between the given values in Table 1 and the simulated values in Table 3 can be compensated if, as in Embodiment 2, a constant calculation parameter F is not used, but rather a variable calculation parameter F (V _j ). The calculation rule T (V _j ) again uses the calculation rule already known from exemplary embodiment 2, but now with the variable calculation parameter F (V _j ):

• • Ausgangspunkt der Iterationsroutine ist der gemäß Ausführungsbeispiel 2 bereits automatisch berechnete Wert F₀ = 8,77, der in 3 ebenfalls als gestrichelte horizontale Linie eingezeichnet ist.
• • Nacheinander wird nunmehr für jedes V_j der variable Berechnungsparameter F(V_j) mit einer konstanten Schrittweite verändert (j ist der Produktparameter, hier mit j ∊ {1, 2, 3, 4}). Die Schrittweite wird hierbei mit Bezug auf den bereits automatisch berechneten Wert F₀ = 8,77 festgelegt, z. B. 1‰ von 8,77, was einer Schrittweite von 0,00877 entsprechen würde.
• • Die Veränderung wird jeweils bei dem F(V_j) durchgeführt, für das die resultierende Abweichung (S_j – P_j) am größten ist. Ist die Abweichung positiv, dann wird der variable Berechnungsparameter F(V_j) mit der festgelegten Schrittweite vergrößert, ansonsten verringert. Dieser Schritt wird solange wiederholt, bis ein vorgegebenes Fehlerkriterium mit Bezug auf die realen Marktdaten S_j und die angepassten simulierten Marktdaten P_j minimal wird. Insbesondere kommen auch hier wieder die bereits beschriebenen Fehlerkriterien in Betracht.
• • Gegebenenfalls kann die Schrittweite in Abhängigkeit der resultierenden Abweichung (S_j – P_j) schrittweise verringert werden. Die Iterationsroutine bricht dann ab, wenn die resultierende Abweichung (S_j – P_j) innerhalb der vorgegebenen Auflösung den Wert 0 annimmt.

In the following, an iteration routine is described for the automatic determination of the variable calculation parameter F (V _j ) to the value sequence F ₀ (V _j ), which runs with high security and which guarantees converges. 3 shows the course of the value sequence F ₀ (V _j ) after performing this iteration routine. It is appreciated that the calculation parameters F (V _j) in accordance with this sequence of values F ₀ (V _j) for different values V _j assumes different values F.

• The starting point of the iteration routine is the value F ₀ = 8.77, which is already calculated automatically in accordance with exemplary embodiment 2 and which is included in 3 also drawn as a dashed horizontal line.
• • Subsequently, for each V _j, the variable calculation parameter F (V _j ) is changed with a constant step size (j is the product parameter, here with j ε {1, 2, 3, 4}). The step size is determined here with reference to the already automatically calculated value F ₀ = 8.77, z. 1 ‰ of 8.77, which would be an increment of 0.00877.
• • The change is made in each case at the F (V _j ) for which the resulting deviation (S _{j -Pj} ) is greatest. If the deviation is positive, the variable calculation parameter F (V _j ) is increased with the specified step size, otherwise reduced. This step is repeated until a predetermined error criterion with respect to the real market data S _j and the adjusted simulated market data P _j becomes minimal. In particular, again the error criteria already described come into consideration here.
• If necessary, the step size can be reduced step by step in dependence on the resulting deviation (S _j -P _j ). The iteration routine then stops when the resulting deviation (S _j -P _j ) within the predetermined resolution takes the value 0.

In addition, another condition must be satisfied for the product F (V _j ) · (V _j ). This product must be monotonically increasing depending on the simulated market data V _j . If this condition were not fulfilled, then this could lead to the calculation rule T (V _j ) calculating rising simulated simulated market data P _j from increasing simulated market data V _j . 4 shows the course of the product F (V _j ) * (V _j ) for the values of the calculation parameter F (V _j ) according to 3 , It can be seen that the gradient condition is fulfilled here.

With the values off 3 the adjusted simulated market data P _j = T (V _j ) in a rounding to 3 digits now calculate to:

A comparison with the values S _j from Table 1 shows no deviation. The error between the buyer model and the real market data S _j has thus been reduced to 0 within the illustrated accuracy.

In the optimization of the value sequence F ₀ (V _j ), however, in contrast to the optimization according to embodiment 2, there are in principle several solutions. Decisive in the optimization is therefore here the starting point due to the already automatically calculated value F ₀ = 8.77, which ensures that the iteration routine does not converge to a sub-optimal solution that delivers a realistic change in the input parameters no realistic market data P _j more. Incidentally, the deviation between the optimized value sequence F ₀ (V _j ) and the already automatically calculated value F ₀ = 8.77 can also be taken as a further error criterion. When finding several solutions can be required as an additional constraint that even this error criterion is minimal.

Embodiment 4

After the automatic determination of the calculation parameter F or of the variable calculation parameter F (V _j ), the values in the matrix X or in the matrix W can be selectively changed in order to further investigate the effects of certain product and / or market changes on the market data to be able to.

Starting from the embodiment 2, certain product changes to product 1 are now further investigated in this embodiment. The calculation block with the adaptation parameter F ₀ = 8.77 remains the same 2 further interconnected with the Constant Decision model. The goal for product 1 should be to improve on the product position "functions" by one rank. Products 1 and 3 thus exchange their ranking positions on the "Functions" product position, so that the following input values for the Constant Decision model now result: Table 4 Market segment 4 drills X position strength W buyer interest 1 2 3 4 price 0.2 0.1 0.3 0.4 0.3 features → 0.3 0.4 → 0.2 0.1 0.5 quality 0.4 0.3 0.2 0.1 0.2

The values of the buyer interest remain unchanged. For the changed position strength values, the following applies in matrix form:

The multiplication of the matrix W with the matrix X yields:

The column sums of the matrix R yield the simulated market data V _j according to the constant-decision model: V _j = (0.29 0.29 0.23 0.19)

The adjusted simulated market data is now:

From the known unit prices, the corresponding market shares can be calculated again.

For comparison purposes, all given and calculated values are summarized in the following table: TABLE 5 Market segment with 4 drills product 1 2 3 4 known number of pieces S _j 0.224 0.361 0.277 0.138 known unit prices c _j in EUR 130.00 170.00 110.00 90.00 known market shares Y _j 0.22 0.46 0.23 0.09 adapted simulated number of pieces P _j .2167 .3359 .3077 0.1397 simulated market shares Y _j , calculated from P _j 0.21 0.43 0.26 0.10 adapted simulated number of pieces P _j with an improvement of product 1 .3326 .3326 0.1965 .1383 simulated market shares Y _j , calculated from P _j 0.32 0.42 0.16 0.10

Embodiment 5

The same changes as in Embodiment 4 will now be made starting from Embodiment 3 with the variable calculation parameter F (V _j ) according to FIG 3 calculated.

The column sums of the matrix R result according to the constant decision model initially the same simulated market data V as in embodiment 4: V _j = (0.29 0.29 0.23 0.19)

The simulated market data V ₁ and V ₃ for products 1 and 3 have changed compared to the original market situation. V ₁ now has the same value as V ₂ , so that the same calculation parameter F (V ₁ ) = F ₂ = 8.91 can be used for both simulated market data. For V ₃ = 0.23, however, the new calculation parameter F (V ₃ ) must be calculated by interpolation. For the interpolation, all known interpolation methods can be used, for. If, for example, a straight line is used as the interpolation between two points, the new calculation parameter F (V ₃ ) is calculated from the values according to FIG 3 with the two-point straight line shape as follows: F (V ₃ ) = 8, 77 9 - 8, 54 / 0.24 - 0.19 · (0.23 - 0.19) + 8.54 = 8.7312

The adjusted simulated market data is thus:

For comparison purposes, all given and calculated values are also summarized in the following table for this exemplary embodiment: TABLE 6 Market segment with 4 drills product 1 2 3 4 known number of pieces S _j 0.224 0.361 0.277 0.138 known unit prices c _j in EUR 130.00 170.00 110.00 90.00 known market shares Y _j 0.22 0.46 0.23 0.09 adapted simulated number of pieces P _j 0.224 0.361 0.277 0.138 simulated market shares Y _j , calculated from P _j 0.22 0.46 0.23 0.09 adapted simulated number of pieces P _j with an improvement of product 1 0.3396 0.3396 .1909 .1299 simulated market shares Y _j , calculated from P _j 0.33 0.43 0.15 0.09

Embodiment 6

In the embodiment 3, the value sequence F ₀ (V _j ) is set with respect to the calculated interpolation points F ₁ = 8.779, F ₂ = 8.910, F ₃ = 8.282 and F ₄ = 8.540 depending on the value V of the simulated market data. The interpolation points are connected or interpolated depending on the value V of the simulated market data.

A variant of the exemplary embodiment 3 consists in determining the value sequence F ₀ (V _j ) with reference to the calculated interpolation points F ₁ = 8.779, F ₂ = 8.910, F ₃ = 8.282 and F ₄ = 8.540 depending on the product parameter j ( here with j ε {1, 2, 3, 4}). 5 shows the corresponding course of the calculation parameter F (V _j ) for the four products j. The calculation of the points F ₁ , F ₂ , F ₃ and F ₄ is identical to the calculation described in Example 3. However, the sequence of values F ₀ (V _j ) is now determined by continuing the corresponding interpolation point F _j for each product j. The value sequence F ₀ (V _j ) thus corresponds to the support points F ₁ , F ₂ , F ₃ and F ₄ already calculated in the exemplary embodiment 3.

With the input parameters according to Table 2, the same adjusted simulated market data P _j = T (V _j ) as in Embodiment 3 initially results, since the interpolation points F ₁ , F ₂ , F ₃ and F ₄ have remained identical. Here, too, there is no deviation from the values S _j from Table 1:

However, there is a difference to the results of embodiment 5, because after the values in the matrix X or in the matrix W have been deliberately changed, the calculation parameter F (V _j ) is no longer determined from the value sequence F ₀ (V _j ) according to 3 but from the sequence of values F ₀ (V _j ) according to 5 , For comparison, therefore, the same changes as in Embodiment 5 are now calculated with the variable calculation parameter F (V _j ) when it is calculated from that according to FIG 5 determined value sequence F ₀ (V _j ) is determined.

The column sums of the matrix R result according to the constant-decision model initially the same simulated market data V as in embodiment 5: V _j = (0.29 0.29 0.23 0.19)

The adjusted simulated market data now calculate to:

In a further variant, the value sequence F ₀ (V _j ) can be determined both as a function of the value V of the simulated market data and as a function of the product parameter j. The sequence of values F ₀ (V _j ) is then a function of the two variables V and j. The calculation of the adjusted simulated market data P _j from the simulated market data V _j proceeds as already described above. For the simulated market data V _j , the respective calculation parameter F (V _j ) from the set value sequence F ₀ (V _j ) is first determined as a function of the value V of the product j. With these calculation parameters F (V _j ), the calculation rule T (V _j ) in the calculation block is then executed. The result is the adapted simulated market data P _j .

Embodiment 7

In Embodiment 2, the values of the matrix X were set by a marketing expert. However, certain input parameters of the matrix X can alternatively be calculated more accurately.

When arithmetic determination of certain values of the matrix X is to be distinguished according to whether the relevant values depend on objective or subjective market criteria. Only values of matrix X that depend on objective market criteria can be calculated.

Subjective market criteria are based on the subjective assessment of the buyer, such. B. the design of a wristwatch. The features and quality features in Table 2 are therefore based on subjective market criteria. In determining the values X, the marketing expert has oriented himself to the question of the probability with which the buyers classify the respective product as the market leader for this product feature.

Objective market criteria are based on numerically known quantities, such. B. the accuracy of a wristwatch. The arithmetical determination of a value X can be based on the question of what size buyers expect on average. In response to this question, the objectively _large quantities c _ij allow an objective average size c _i set:

The product feature "Price" in Table 2 is based on objective market criteria. With the index i = 1, the average price is valid according to the values in Table 1: c ₁ = 133.38 EUR

A position strength x _ij , which depends on objective market criteria, can generally be determined as follows:

und ein zweiter Hilfswert

berechnet.Before the actual value x _{ij is} calculated, first a first auxiliary value is calculated

and a second auxiliary value

calculated.

zwei Fälle unterschieden: Das ”+” gilt, wenn eine Erhöhung von c_ij auch eine Erhöhung von x_ij zur Folge hat, also bei direkter Proportionalität. Als Beispiel wäre hier die Ganggenauigkeit einer Armbanduhr zu nennen. Das ”–” gilt dagegen, wenn eine Erhöhung von c_ij eine Verringerung von x_ij zur Folge hat, also bei umgekehrter Proportionalität Als Beispiel wäre hier der Preis zu nennen.In the first line are used in the calculation of the first auxiliary value

distinguish two cases: The "+" applies if an increase of c _ij also _results in an increase of x _ij , that is, in direct proportionality. As an example, the accuracy of a wristwatch should be mentioned here. On the other hand, the "-" applies if an increase of c _{ij results in} a reduction of x _ij , that is, if the proportionality is reversed. As an example, the price should be mentioned here.

eine Normierung auf die Summe 1.The second line is for the second auxiliary value

a normalization to the sum 1.

. Da α_{x_ist} grundsätzlich auch sehr kleine Werte annehmen kann (d. h. die zahlenmäßig bekannten Größen c_ij unterscheiden sich dann nur geringfügig voneinander), wird der Logit-Faktor F_x begrenzt, beispielsweise auf den maximalen Wert F_x = 4·n. Bei Bedarf kann für den Faktor 4 auch eine andere Zahl gewählt werden.In the third line, the calculation of the values x _ij takes place with a normalization to the relative standard deviation α _{x_soll} . The relative standard deviation α _{x_soll} is determined by the rows of the matrix X, which depend on subjective market criteria. The goal is that all rows of the matrix X have approximately the same relative standard deviation α _{x_soll} . α _{x_act} is the relative standard deviation of the second auxiliary values

, Since α _{x_ ist can in} principle also assume very small values (ie the numerically known quantities c _ij then differ only slightly from one another), the logit factor F _{x is} limited, for example to the maximum value F _x = 4 * n. If necessary, a different number can be selected for the factor 4.

:

With the average unit price of EUR 133.38 and the unit prices according to Table 1, the second auxiliary values apply

:

berechnet sich dabei zu α_{x_ist} = 0,2087_. Die relative Standardabweichung α_{x_soll} in den Zeilen 2 und 3 der Matrix X beträgt außerdem α_{x_soll} = 0,4472. Damit berechnet sich der Logit-Faktor F_x zu:

The relative standard deviation of the second auxiliary values

is calculated as α _{x_ist} = 0.2087 _. The relative standard deviation α _{x_soll} in lines 2 and 3 of the matrix X is also α _{x_soll} = 0.4472. This calculates the logit factor F _x to:

The position strengths on the product position "Price" are calculated as follows after applying the logit transformation with the logit factor F _x = 8.5712:
x ₁₁ = 0.2176 x ₁₂ = 0.1279 x ₁₃ = 0.2840 x ₁₄ = 0.3705

This changes the values in Table 2 as follows: Table 7 Market segment with 4 drills X position strength W buyer interest 1 2 3 4 price .2176 .1279 0.2840 .3705 0.3 features 0.2 0.4 0.3 0.1 0.5 quality 0.4 0.3 0.2 0.1 0.2

The matrix W remains unchanged.

In matrix form, the values of the matrix X are now as follows:

The multiplication of the matrix W with the matrix X yields:

The column sums of the matrix R yield the simulated market data V _j according to the constant-decision model: V _j = (0.2453 0.2984 0.2752 0.1812)

mit f(S_j) = f₃(S_j) und b = 0,6 gemäß Ausführungsbeispiel 1 sowie aufgrund von

berechnet die Iterationsroutine des Berechnungsblocks automatisch den Wert F₀ = 8,19. From the condition

with f (S _j ) = f ₃ (S _j ) and b = 0.6 according to Embodiment 1 as well as due to

the iteration routine of the calculation block automatically calculates the value F ₀ = 8.19.

The iteration routine returns the value 0.00212159 as the minimum value for the error criterion.

The adjusted simulated market data is calculated as:

Tabelle 8 Marktsegment mit 4 Bohrmaschinen Produkt 1 2 3 4 angepasste simulierte Stückzahlenanteile P_j 0,2265 0,3500 0,2895 0,1340 simulierte Marktanteile Y_j, berechnet aus P_j 0,22 0,45 0,24 0,09 From the known unit prices, the corresponding market shares can be calculated according to:

Table 8 Market segment with 4 drills product 1 2 3 4 adapted simulated number of pieces P _j .2265 .3500 .2895 .1340 simulated market shares Y _j , calculated from P _j 0.22 0.45 0.24 0.09

The correspondence between the values in Table 8 and in Table 1 is now even better than between the values in Table 3 and in Table 1, since the input parameters for the price are less error-prone.

Embodiment 8

For the buyer model according to Embodiment 7 with interconnected calculation block shows 6 the course of the simulated market shares after the unit prices have been selectively changed one after the other. This targeted change has now become possible due to the mathematical determination of the values x _1j for the product _feature price. This gives you the unit price gradient for each product. Only for product 4, the unit price gradient therefore has a positive slope, so that only for product 4, the market share can be increased by increasing the unit price.

• • Für jedes Produkt j wurde der entsprechende Wert x₁ schrittweise verändert. Für diese Veränderung wurde der entsprechende Stückpreis c_1j durch Interpolation ermittelt. Danach wurde die Zeilensumme der Zeile 1 der Matrix X mit dem veränderten Wert x_1j wieder auf den Wert 1 normiert.
• • Für jede schrittweise Veränderung des besagten Wertes x_1j wurde nach der Zeilennormierung der resultierende Marktanteil Y_j berechnet.

The unit price gradients according to 6 were determined as follows:

• • For each product j, the corresponding value x _{1 has been} changed step by step. For this change, the corresponding unit price c _1j was determined by interpolation. Thereafter, the row sum of the row 1 of the matrix X with the changed value x _{1j was normalized} again to the value 1.
• • For each stepwise change of said value x _1j , the resulting market share Y _{j was} calculated after the line _{normalization} .

In addition, when changing the values x _1j in this _exemplary embodiment, the values w _{i were also} adaptively changed. For price increases, the value w _{1 was} increased and the values w ₂ and w _{3 were} reduced accordingly. In the case of price reductions, on the other hand, the value w _{1 was} reduced and the values w ₂ and w ₃ increased accordingly. This additional measure is recommended because the buyer's interest in the product feature price increases with price increases and decreases accordingly with price reductions. Incidentally, this measure can always be used whenever certain input parameters of the Buyer model after the interconnection of the calculation block to be selectively changed and thereby also the average unit price c ₁ changed.

A variant of the method according to the invention is present when the structure of the purchaser model is changed during a targeted change of the input parameters. For example, it is conceivable that a product is omitted and thus changes the number of columns of the matrix X and thus also the matrix R. The inventive interconnection of the calculation block with the output of the buyer model is then still possible if the automatically determined calculation parameter is changed accordingly:

A further variant of the method according to the invention is that the calculation block is not followed by the buyer model, but rather that one or more calculation blocks are used in the buyer model where percentage values with a total of 100% are present. In the Constant Decision model, this applies to the values of the matrix W and to the row values of the matrix X. In the adaptation, all values of the matrix W and / or all row values of the matrix X could be assigned a respective calculation block, whereby all calculation blocks act on the values during a iterative adaptation by a common adaptation parameter F. Otherwise, the corresponding iteration routine may automatically run with the same error criteria as described above.

QUOTES INCLUDE IN THE DESCRIPTION

This list of the documents listed by the applicant has been generated automatically and is included solely for the better information of the reader. The list is not part of the German patent or utility model application. The DPMA assumes no liability for any errors or omissions.

Cited non-patent literature

• Gary L. Lilien, Arvind Rangaswamy, Marketing Engineering, 2nd Edition, Trafford Publishing, 2004 [0002]
• Gary L. Linien, Arvind Rangaswamy, Marketing Engineering, 2nd Edition, Trafford Publishing, 2004, p. 46. [0015]

Claims (19)

Computer-implementable method for automatic adaptation of market data V _j simulated with erroneous survey values to real market data S _j with the number n, wherein the error-prone survey values are input parameters of a buyer model and the simulated market data V _j with the number n are calculated at the output of the buyer model, characterized in that a calculation block with a calculation rule T (V _j ) is provided, wherein the calculation block calculates simulated market data P _j = T (V _j ) from the simulated market data V _j with the number n, the calculation rule T (V _j ) of a calculation parameter F (V _j ) and the calculation block automatically sets the calculation parameter F (V _j ) to a value sequence F ₀ (V _j ), so that a predetermined error criterion with respect to the real market data S _j and the adjusted simulated market data P _{j is} minimal and the input of the calculation block while keeping the automatically determined value sequence F ₀ (V _j ) is connected to the output of the purchaser model. Computer-implementable method according to claim 1, characterized in that the calculation block computes the adjusted simulated market data P _j according to

wherein the calculation parameter F (V _j ) assumes the same value F for all V _j . Computer-implementable method according to claim 2, characterized in that the automatically fixed value sequence F ₀ (V _j ) for all V _{j takes} the same value F ₀ , wherein the automatic determination of the value F _{0 is carried out} by an iteration routine, in which the error criterion by smaller expectant changes .DELTA.F to the value F is gradually minimized, the iteration routine aborts when the changes .DELTA.F have fallen below a predetermined resolution. Computer-implementable method according to claim 1, characterized in that the calculation block computes the adjusted simulated market data P _j according to

wherein the calculation parameter F (V _j ) is set to a value sequence F ₀ (V _j ) containing interpolation points F _j for which, within a given resolution, the resulting deviation (S _j -P _j ) becomes zero. Computer-implementable method according to claim 4, characterized in that the value sequence F ₀ (V _j ) outside the interpolation points F _j is determined by the fact that the interpolation points F _j are interconnected or interpolated as a function of the value V of the simulated market data. Computer-implementable method according to claim 4, characterized in that the value sequence F ₀ (V _j ) outside the interpolation points F _j is determined by the fact that for each parameter j the corresponding interpolation point F _{j is} constantly continued and thus the same for the corresponding market data value V _j Takes value F. Computer-implementable method according to one of claims 1 to 6, characterized in that the error criterion is specified by the sum of the error squares according to

Computer-implementable method according to one of claims 1 to 6, characterized in that the error criterion is specified by a sum of normalized error squares according to

Computer-implementable method according to one of claims 1 to 8, characterized in that certain input parameters of the purchaser model are selectively changed after the interconnection of the calculation block. A computer program product for implementing and executing the method of any of claims 1-9 on a computer. A computer for automatically matching market data simulated with erroneous polling values, _Vj, to real market data _{Sj of} the number n, wherein the real market data S _{j is} stored in the computer, wherein the polling polling values are stored in the computer and input values of one implemented on the computer Representing buyer model and the computer at the output of the buyer model, the simulated market data V _j with the number n calculates and stores, characterized in that the computer has a calculation block with a calculation rule T (V _j ), wherein the calculation block from the simulated market data V _j adapted simulated market data P _j = T (V _j) with the number automatically n calculates and stores, wherein the calculation rule (V _j) is dependent on a calculation parameter F (V _j) T, and the calculation block to calculate parameters F (V _j) to a sequence of values F ₀ (V _j ) so that a given error criterion m is minimal with respect to the real market data S _j and the adjusted simulated market data P _j , and wherein the computer has an interface through which the input of the calculation block maintains the automatically determined value sequence F ₀ (V _j ) with the output of the buyer model is interconnectable. Computer according to claim 11, characterized in that the calculation block computes the adjusted simulated market data P _{j in} accordance with

wherein the calculation parameter F (V _j ) assumes the same value F for all V _j . A computer according to claim 12, characterized in that the automatically determined value sequence F ₀ (V _j ) assumes the same value F ₀ for all V _j , wherein the automatic determination of the value F _{0 is carried out} by an iteration routine in which the error criterion becomes smaller Changes ΔF is minimized at the value F step by step, wherein the iteration routine aborts when the changes .DELTA.F have fallen below a predetermined resolution. Computer according to claim 11, characterized in that the calculation block computes the adjusted simulated market data P _{j in} accordance with

wherein the calculation parameter F (V _j ) is set to a value sequence F ₀ (V _j ) containing interpolation points F _j for which, within a given resolution, the resulting deviation (S _j -P _j ) becomes zero. Computer according to Claim 14, characterized in that the value sequence F ₀ (V _j ) outside the interpolation points F _j is determined by interlinking or interpolating the interpolation points F _j as a function of the value V of the simulated market data. Computer according to claim 14, characterized in that the value sequence F ₀ (V _j ) outside of the interpolation points F _j is determined by the fact that for each parameter j the corresponding interpolation point F _{j is} constantly continued and thus for the corresponding market data value V _j the same value F accepts. Computer according to one of claims 11 to 16, characterized in that the error criterion is given by the sum of the error squares according to

Computer according to one of claims 11 to 16, characterized in that the error criterion is given by a sum of normalized error squares according to

Computer according to one of claims 11 to 18, characterized in that certain input parameters of the purchaser model are selectively changed after the interconnection of the calculation block.

Images