Effective position of European firms in the face of Monetary Integration using Kohonen’s SOFM
Raquel Flórez López
Department of Economics and Business Administration
University of León (Spain)
Campus de Vegazana, s/n - 24071 León (Spain)
Phone: +34-987-291734
E-mail:
1
Abstract
The Economic and Monetary Union (EMU) is the culmination of the European integration process as far as the financial perspective is concerned, which main aim is the implantation of a one and only currency for all the member states included in this integration project which is going to take place in early 2002. To decide the relation of countries included in this phase it was established a relation of macroeconomics regulations known as Convergence Criteria or Maastricht Criteria which must be fulfilled to guarantee the economic convergence among countries. Nevertheless, these criteria are not enough to assure the effective convergence among countries as far as enterprises is concerned, due to internal national differences in microeconomics structures, which affect to competitiveness among firms and could distort a lot the effect of the Union in favour of some countries and against others.
The additional use of Artificial Neural Networks together to classical analysis techniques, specifically the employment of Self-Adaptive Models based in Kohonen’s proposal, makes easier to analyse these microeconomics differences, getting a visual image of particular positions through topological two-dimensional maps.
Keywords.Economic and Monetary Union, Artificial Neural Networks, Kohonen's SOFM, financial and economical analysis.
- The Economic and Monetary Union Process
The European Economic and Monetary Union (EMU) is an old aim of European countries in order to get a real stability monetary zone in Europe. This process, that began in early 70’s, was finally culminated in 1990 with the Maastricht Treaty, where it was accepted the ‘Delors Plan’ to facilitate the integration process, being its three main phases the following ones: Preparation stage (1-7-1990 to 1-12-1993), characterised by the establishment of the Convergence Program, based on five macroeconomics Convergence Criteria which should be satisfied for the countries that desired to participate in the launch stage: inflation, exchange rate, National Deficit, National Debt and the interest rate. Consolidation stage (1-1-1994 to 1-12-1998), transitory phase in order to increase the convergence among countries in terms of their economic and monetary policies. Launch stage (1-1-1999 to 2002), with the establishment of the final fixed exchange rates, the implantation of the Euro as the European Currency, and the definition of a common Monetary Policy for the European Central Bank, created in this phase together to the European System of Central Banks (ESCB), organism responsible of the common monetary policy.
While the two early phases were obligatory for all the member states, the access to the third stage was conditioned to the accomplished of the Convergence Criteria. In addition, United Kingdom and Denmark were authorized to voluntary excluded themselves of this third phase (which they do) and Sweden could internally decide to participate or not in the Launch stage (which declined too). Finally eleven countries primary accessed to the third phase: Austria, Belgium, Finland, France, Germany, Ireland, Italy, Luxembourg, the Netherlands, Portugal and Spain (Greece was included in 2001).
Even when this process presents many microeconomics advantages, however it has many disadvantages also, due to the five Convergence Criteria’s main aim was to guarantee the similarity of European countries in macroeconomics terms, not considering the important microeconomics differences in terms of costs, benefits, profitability and patrimonial situation that will probably increase due to this harmonisation process. The objective of this paper is to investigate the economic and financial relative position of different countries’ firms to face these challenges, using both the classical analysis based on ratios and the Machine Learning technique known as Artificial Neural Networks (specifically Kohonen’s self organizing model).
2. Self-Organizing Feature Map
Artificial Neural Networks (ANNs) constitute a Machine Learning paradigm that appeared like some attempt to establish mathematical formulations about brain structure, characterised by learning through the experience and the knowledge extracting from multiple different events.The architectures used in ANNs can be divided into three categories, from the point of view of data flow direction: feedforward networks, wherethe information flows in an only direction, feedback networks, where data can flow in different directions between layers and competitive, unsupervised or self-organizing models, based on competition among neighbouring cells in the net through mutual lateral interactions, so that neurons evolve to specific detectors of different signal patterns. The Self-Organizing Feature Map (SOFM) proposed by professor Kohonen in early 80’s belong to this third category.
The model is basically composed by two layers; the first one, or input layer, is formed for m cells, one for each system input variable, which task is to pick up the data from the environment; the second one, or output layer, usually presents a rectangular feature and makes all the information processing (Figure1).
Figure 1
The SOFM model presents two different stages:
1.- Learning or Training Stage: In the learning stage the network establishes the categories which will be used in the operation stage to classify new data. The training process is characterised by the successive presentation to the system of different patterns; each time an individual x(t) is treated by the system, the net obtains a “winner cell” m*ij(t) (the nearest neuron to the pattern) which will modify its weight vector together to other neurones situated in its neighbourhood (calculated through a function h(i-c,t) that determinates the ‘neighbourhood radio’ to actualise weights) in the following way:
mijk(t+1)=mijk(t)+(t).h(i-c,t).(xk(t)-wijk(t)), 0<<1.
The process is repeated for many patterns (when the entire patterns have been presented to the network, usually in a random way to avoid slanted systems, it is called on “epoch” of learning), so different weight vectors approach the probability density function of the inputs, in a process, that has got quite similarities with the clustering method. The training process used to finish after some pre-fixed steps or epochs (for example, a ‘thumb rule’ stops the system at 100.000 steps). In addition, it is possible to perform a second training stage, known as “fine tuning” to get a better fit of the weight vectors. In this moment the learning rate is very small (near 0.01) and the neighbour radio is fixed to 1 to obtain the adjustment of each output layer’s weight vector.
2.- Operation or Working Stage: After the learning stage the working stage begins, where the weight vectors remain fixed. The operation phase is relatively simple, generating an only one output from an input vector. This output is obtained through the parallel calculus for each neurone in the output layer of the distance between the input data and its weight vector. The neurone with the maximum similarity is established as winner, and the input vector is assigned to this cell. In that way, each cell acts like a specific feature detector, where the winner points the feature or patron detected in the input data.
3. Empirical analysis
The database employed in the study of European enterprises as far as their relative competitive position is concerned has been obtained from the ‘BACH Project Database’ (Bank for the Accounts of Companies Harmonized), which collects in a harmonised way the accounting states of non-financial companies in 13 countries broken down by major activity sector and size. The presented study has used 1993 to 1998 financial data from the nine countries integrated both in the Launch Stage of EMU and in BACH database: Germany, France, Italy, Spain, Belgium, the Netherlands, Portugal, Austria and Finland, as far as the overall manufacturing industry sector is concerned, reducing the effect of different enterprises’ size through the employment of ratios’ technique. The period considered covers near all the Preparation and Consolidation Stage and the first year of the Launch Stage, which is important to analyse the evolution of different European microeconomics position during the global EMU period. In order to evaluate the competitive position of European enterprises, the study was concentrated in their patrimonial equilibrium and their cost and profitability situation.
These items were analysed by means of seven financial ratios, grouped in two different categories, using a 2-dimensional SOFM model to get an overall view of firms’ relative situation in terms of these classes. In that way, there are others techniques that could be employed to study these data, like multidimensional scales, discriminant analysis or cluster analysis, but these methods have some disadvantages that should not be forgotten, like the production of very overlapping segments in presence of outliers or the fulfillment of very restrictive data hypothesis (normality, homocedasticity). In addition, none of these techniques lets to obtain an overall 2-dimensional map. SOFM model avoids these difficulties, but requires a higher analysis of internal parameter, and is not mathematically optimal, but based on pragmatic learning techniques. The employment of SOFMs involved the definition of some critical variables as far as both neural architecture and learning process is concerned. In that way, inputs variables were financial quotients and outputs variables were the analysed nine countries and six years (54 records), that were positioned using their relative financial similitude.
The number of output units was obtained using ‘Sammon’s mapping’, and initial weights were randomly selected in the [0, 0.01] interval [(Kohonen, 1997), (Martín y Sanz, 1997)]; it was employed a “two-phases” learning: the “rough learning stage” (stopped about 100.000 iterations) and the “fine adjustment stage” (conformed of about 1.000 iterations). Both, neighbourhood ratio and learning rate were progressively decreased to improve results.
3.1. Economic analysis (I): Cost analysis
The objective or main aim of the economic analysis is to study the profitability of firms (capacity to generate profits), so the needed information to develop this investigation comes basically from the Profit and Loss State, using some magnitudes from the Balance-sheet too. To analyse this aspect, it is very useful to study two different situations:
- Analysis of the cost structure in the enterprise.
- Analysis of its effective profitability.
As far as the analysis of cost structure is concerned, three different quotients were considered:
1.Staff costs/AAV, which could be interpreted like the labour unitary cost, is one of the most employed cost indicators used to measure firms’ competitiveness. The Adjusted Added Value (AAV) included in the denominator represents the value generated for the enterprise in its productive operations.
2.Cost of materials and consumables/Net turnover, that represents the percentage of turnover employed in tangible consumes needed to develop the production cycle.
3.Interest and similar charges/AAV, that informs about the portion of AAV used to reward external creditors of the enterprise. In that way, the financial policy of the firm will determine in a very important way the evolution of this indicator.
Analysing the BACH database, it was found that the third ratio could not be calculated for Netherlands (period 1993-1994) and Austria (period 1993-1998), so it was not introduced in Kohonen’s map, having been evaluated isolated. This indicator showed that the financial cost have decreased in all countries treated, being finally situated in its majority in an interval of 7-9%. The highest decreases are experimented by Finland, Spain and Portugal, which passed of ratios plus than 20% to quotients between 5-10%.
The Kohonen’s map finally obtained, considering both previous quotients is shown in Figure 2.
Fi-97 Fi-98 / Fi-95 / Fi-93 Fi-96 / A-94 / A-93 A-96 / G-94 / G-93Fi-94 / A-95 / A-97 / G-97 G-98 / G-95 G-96
N-28 P-95 / P-94 / A-98
P-97 / N-95 / P-93
N-98 N-97 P-98 / P-96 / N-94 / N-93 / F-95 / F-94
I-95 / S-95 / F-98 / F-97 / F-93 F-96
S-98 / S-97 / B-98 B-97 / I-94 B-96 / I-93 S-94
I-97 / I-98 / I-96 / S-96 B-95 / B-94 / B-93 / S-93
Figure 2
(G=Germany, F=France, I=Italy, S=Spain, N= the Netherlands, B=Belgium, P=Portugal, A=Austria, Fi=Finland)
The analysis of Kohonen’s weights map lets observe five different topological regions:
1. Finland constitutes an isolate region, characterised by a low labour rate (near 55-60%) and a low material rate (about 55-60%) which confers it a very good position for the EURO introduction stage.
2. Germany and Austria conform another separated region, characterised for a very high labour cost (about 70-80%) but quite low material rate (near 55%). In conclusion, in spite of their good cost of material’s position, these countries could have some problems to compete with nations with less labour costs.
3. France forms a separated region, with medium labour rate (65-70%), decreasing the last years of the period and medium-high material ratio (upper than 70%). So this country is situated in a ‘medium term’ even when it must watch its material costs.
4. Portugal and the Netherlands constitute another region, with low labour rate (about 55-65%) and high material cost (about 70-75%). These countries are near to France position, so they could be its main competitor as far as these variables are concerned.
5. Finally, Spain, Italy and Belgium constitute the last region, characterised by an important movement in the period from labour rates of about 70-80% to 60% ratios, and from material cost from 70% to upper than 75% (as a whole). These countries present the lowest labour costs, which can be its main competitive advantage, but its material cost are very high and increasing for the last year, so they must be careful in this point for not to lose the previous advantage.
In conclusion, there are quite important differences in cost-ratios among European countries, specially between ‘north-countries’ and ‘south-countries’ (with the exception of Portugal and the Netherlands, situated in a intermediate position and very near one from another and Belgium, integrated in the “south group”).
3.2.Economic analysis (II): Profitability analysis
The profitability, defined as the firm's capacity to generate profits, results basic to face the increase of competence. The main variable used to illustrate this analysis, together to the financial leverage, is the economic profitability, calculated as ‘Profit before financial result and taxes/Net Assets’. This item, which shows the efficiency of the business in the inversion of its resources, can be divided in two different magnitudes, the ‘margin effect’ and the ‘rotation effect’. In that way, this magnitude is going to be studied through four different variables:
- Economic profitability (1): Known as the ‘margin effect’ of the economic profitability, it is calculated like ‘Profit before financial result and taxes/Net turnover’. A high ratio will indicate a high profit value per each sold product.
- Economic profitability (2): This is the ‘rotation effect’, defined as ‘Net turnover/Net Assets’, and indicates the number of times in which the inverted money inverts is sold.
- Financial profitability: This ratio informs about the profitability gets for the business’ owners, so this magnitude has into account the composition of the Liabilities of the firm. This variable is calculated like ‘Profit before taxes/Own resources (capital and reserves in BACH database)’.
- Cost of payable liabilities: This magnitude informs about the specific cost of the non-owned liabilities. Its value depends both of the interest rate and of the financial politics of the enterprise. Due to the interest rate has been one of the Convergence Criteria in this period, these ratios show basically the financial politics of the firm.
It may be commented that this analysis could not be done for the Netherlands (1993-1994), Finland (1993) and Austria (1993-1998), so the obtained map does not collect these countries and years. The Kohonen’s map finally obtained is shown in Figure 3.
Fi-97 Fi-98 / N-98 / N-96 N-97 / I-98 / I-97 / I-96 / I-94N-95 / B-96
Fi-94 Fi-95 Fi-96 / B-98 / B-97 / F-94
P-97 / P-98 / I-95 / F-95 / F-96
P-95 / B-95 / S-97 / S-98
P-93 / P-94 / P-96 / B-94 / S-95 S-96 / G-98
S-94 / F-97 / G-97
S-93 / I-93 B-93 / G-93 F-93 / G-94 / G-95 G-96 / F-98
Figure 3
The previous figure allows to observe that this second economic analysis is the one with less evident differences inter countries, that is to say, the regions formed are more difficult to identify. Anyway, it is able to distinct six regions:
1. Finland constitutes again a specific region, characterised for the highest ‘margin ratio’ (plus than 10%) but a relative quite low ‘rotation effect’ (about 0.7 times), so the economic profitability obtained can be categorised as medium or medium-high. Anyway its financial profitability is the highest (plus than 20%), and its financial cost is low-medium (between 4.2% and 5.7%), so its overall profitability situation can be comparatively defined as very good all over the period.
2. The Netherlands constitutes a different zone too, very near to Finland. The main differences are a lower ‘margin effect’ (7-8.5%) and a higher ‘rotation effect (approximately 0.75 times), so its general economic profitability is lower that Finland’s one. Its financial profitability is lower too (about 14%) and its financial cost ratio is a little lower too (near to 3.5%). In general its economic situation as far as profitability is concerned is quite good and stable along time.
3. Portugal conforms a new group (even when the position for 1993 is a little far from the others years) characterised by a medium-high ‘margin effect’ (5-7%) and ‘rotation effect’ (near 1). So its economic profitability is quite similar to Finland’s one, but obtained due to different factors. The financial profitability is positive but a little low (7-10%) and its financial cost ratio is the highest of all countries, even when it has decreases for 12% to 6%. In that case Portugal presents a general good profitability situation, but it must watch its financial policy to reduce the financial costs.
4. Italy constitutes in the last years a different region (together with a specific situation of Belgium in 1996), characterised by medium low ‘margin effect’ ratio (progressively increased) and medium-high ‘rotation effect’ (a little upper than 1). Its financial profitability as increased from 10.47% in 1996 to 14.58% in 1998, so it is quite good situated. Finally its financial costs have decreased until 3.46% in 1996 which is a very good position. In that way, Italy is quite good situated in this sort of magnitudes, showing a good evolution in the analysed period.