Regional Labour Markets in Germany: Statistical Analysis of Spatio-Temporal Disparities and Network Structures

4 Chapter 1 
 
In this framework, the analysis of regional labour markets is also of great importance for 
their financial and socio-economic implications. Localized top-to-bottom labour policies may 
address more efficiently the specific problems of single regions or areas. Moreover, the 
funding for unemployment benefits or employment programmes is distributed, in countries 
such as Germany, on a regional basis. Being able to effectively forecast labour market 
aggregates, such as unemployment levels, is therefore critical. 
These aspects assume even greater relevance if we consider the wide disparities in the 
performance of regional labour markets that many countries experience (see, for instance, 
Elhorst 2003; Bayer and Juessen 2007). We can think of the differences between the northern 
and southern areas in Italy and between West and East Germany, or between urbanized and 
rural areas. Regional disparities are a conspicuous cost for the national economy: for instance, 
because of the welfare policies necessary to support underperforming areas, and because the 
phenomenon of long-term unemployment – frequent in high-unemployment areas – is often 
associated with problematic socio-political aspects (for example, criminality or political 
sclerosis), which add up to the aforementioned costs (see, for instance, Gilles 1998). Regional 
disparities may also slow down the economic performance of the country as a whole, because 
of the consequent inefficient allocation of resources. And, finally, a high heterogeneity among 
regions makes it difficult to provide accurate estimates of their reactions to national trends or 
shocks (Blanchard 2003). In such a scenario, the analysis of the evolution of regional 
disparities, in particular in the short run, is problematic, as many factors which may determine 
heterogeneity deserve thorough consideration. Just a few examples are: the concentration of 
capital due to positive agglomeration externalities; constraints in land use regulations; varying 
quality of infrastructure.  
In neoclassical economics, disparities among regions (for example, in per capita income) 
are expected to decline in the long run, as capital and labour tend to move towards lower- and 
higher-wage regions, respectively, until the equalization of the two factors’ productivity in all 
regions is reached (Armstrong and Taylor 2000). In addition, underperforming regions have 
the possibility of catching up with the richer ones in technology. However, in the short run, 
regional disparities do seem rather persistent (see, for example, Elhorst 2003). 
The present study is concerned with the statistical analysis of the economic indicators 
underlying the aforementioned disparities. Employing disaggregated data sets on German 
regions for our case studies, we analyse and model the heterogeneity in regional labour 
markets. In this context, the relevance of the space-time components is taken into account 
according to different methodological approaches. In addition, we examine regional labour 
mobility – by means of network theory – in order to better interpret the dynamic patterns of 
the dominant and marginal areas. 
The theoretical-methodological framework – and the related empirical research questions 
– supporting the above research agenda are outlined in the next sections. 
 
 Introduction: Theoretical and Policy Aspects 5 
 
1.2  Labour Markets and Economic Output 
 
The performance of labour markets depends on a number of factors (characteristics of the 
labour force, efficiency of the labour demand/supply match, mobility of labour and capital, 
and so on). Microeconomic processes based on individual utility functions, preferences and 
constraints (such as residential location and mobility choices or employment regulations) have 
been – and still are – widely studied from different perspectives (see, for example, White 
1977; Boyce et al. 1988; van Ommeren et al. 1999a,b, 2000). These micro-mechanisms 
determine the single individual’s choices and are at the basis of the aggregate economic 
results that we observe at the regional (and national) level (such as employment levels or 
commuting flows). The analysis of such phenomena is undoubtedly a sine qua non for 
understanding the inner functioning of labour markets (see, for instance, Fischer and Nijkamp 
1987; Topel 1994). On the other hand, the statistical modelling – let alone forecasting – of 
this type of economic process requires extensive data. 
Therefore, we may also look at aggregate labour market outcomes (such as regional 
(un)employment) from an alternative – yet complementary – perspective. In the 1960s, Arthur 
Okun described and interpreted a recurring empirical finding, according to which changes in 
unemployment are related to the growth of the GDP. In other words, this relationship, which 
is now known as Okun’s law (Okun 1970; Prachowny 1993),
1
 links economic output and 
unemployment. In more detail, it is suggested that the growth rate of per capita GDP is 
negatively correlated with simultaneous variations in unemployment. The actual extent to 
which GDP affects unemployment has been widely discussed in the literature (Paldam 1987; 
Prachowny 1993). Not only unemployment, but also employment variations are related to the 
aforementioned relationship. In fact, an increase in production levels corresponds – in the 
short run – to an increase in labour demand. This increase leads to higher employment levels 
and to a consequent decrease in the unemployment rate, since, assuming a conventional 
labour/capital production function, firms adjust to the need for increased production by hiring 
more personnel. 
The short-run increase of the labour factor in production is expected, because the 
additional adjustments to increased production need are long-run phenomena. Over a longer 
period, however, improvements in labour productivity are sought, by augmenting the share of 
capital in the production factors. With regard to labour, adjustments to increased demand may 
ultimately generate migration phenomena (depending on the efficiency of the real estate 
market). These medium- to long-term effects are not analysed in this study, as additional data, 
theories and analytical tools would be needed. 
We choose therefore to focus on short-term economic adjustments. At the national level, 
these adjustments – the effect of economic output variations on the labour market – may be 
                                                 
1
  Prachowny (1993) actually elevates Okun’s ‘law’ to the level of ‘theory’, providing extended theoretical 
discussion and empirical tests on the validity of Okun’s hypotheses. 
6 Chapter 1 
 
influenced by the business cycle, together with demographic and institutional factors (Kosfeld 
and Dreger 2006). The need for additional employees can then be absorbed not only by the 
unemployed, but also, for instance, by an increase in participation rates, as discouraged 
workers resume their search for a job. Similarly, in a scenario of a growing labour force, the 
decrease in the unemployment rate might be less than the related increase in employment. 
At the regional level, the above relations are further complicated by spatial matters. As 
regions are small open economies, they are not self-contained (which is most likely the case 
of nations), but are highly interactive. The aforementioned demographic phenomena seen for 
the national case are still relevant, but with the additional element of the interaction between 
regions. In this framework, one of the effects of an increase in labour demand in a single 
region is the increased mobility – towards that region – of the (potential) workers residing in 
neighbouring regions (that is, increased incoming commuting). Similarly, in a downward 
period, workers who are laid off may seek work opportunities in neighbouring regions. 
Additionally, regions may show consistently different levels (or growth rates) of 
economic output. These regional differentials in output are – following Okun’s law –reflected 
in the regional labour market aggregates; that is, employment and unemployment. On the 
basis of the above discussion, we can stress that the observation of such aggregates provides 
us ‘with a signal of where the [regional] economy stands’ (Blanchard 2003, p. 29). In other 
words, changes in regional labour markets, such as an increase in unemployment rates, can be 
interpreted as an indication (a proxy) of the changes in the levels of regional economic 
activity (for example, per capita regional GDP). Evidence of the existence of the relationship 
between per capita regional GDP and unemployment rates can found in the literature (see, for 
example, European Commission 1996). On the other hand, others are more critical of the 
consistency of this relationship over time, because of concerns about the stationarity of the 
series observed (Elhorst 2003). These concerns are, however, related to full employment 
conditions implicit in Okun’s law and can be considered to be particularly relevant in an 
unemployment/regional GDP regression framework and in the medium to long run. 
 
1.3  Regional Interactions and Persistence of Heterogeneity 
 
As mentioned above, in the short run the main adjustments to changes in the level of 
economic activity arise in labour demand. These changes in output levels are due to 
increased/decreased demand for the goods and services produced in the area concerned. A 
distinction should be made at this point, as to what is the final destination of the output. It is 
safe to assume that a share – the extent of which varies from case to case – of the production 
is consumed within the domestic market, while the remaining quantity of goods/services 
produced is destined for export. 
Setting aside the internal consumption of goods, which depends largely on population size 
and income levels, we can consider two examples of the trade of goods and services. At the 
 Introduction: Theoretical and Policy Aspects 7 
 
national level, the exchange of output will happen over country boundaries. The extent of the 
cross-country trade and, in particular, the finding of a positive or negative balance show 
whether a country is an importer or an exporter or resources. For instance, the USA has had a 
negative international trade balance for more than two decades, while smaller countries such 
as the Netherlands tend to have a positive trade balance (CBS 2007). More generally, with 
some notable exceptions such as the USA, richer countries will have positive trade balances, 
and poorer countries will have negative ones. In principle, the same can be expected on a 
regional basis. On the other hand, the analysis of regional trade poses additional obstacles 
than in the case of nations, the major one being the lack of official statistics measuring the 
flows of goods and services between regions (Armstrong and Taylor 2000; Polenske and 
Hewings 2004). The relevance of this aspect appears to be greater if we consider that regions 
are more open systems than nations. In open systems, interregional trade can be expected to 
absorb a larger share of the total output; more generally, interactions between regions have a 
greater role in the socio-economic development of each single region. It has been observed 
that only geographically small countries, such as the Netherlands, rely on trade as much as 
single regions within a larger country, such as the UK. 
The nature and extent of the economic flows between regions is important in our context 
for two main reasons: 
 
(a) The aforementioned Okun’s law refers to closed economic systems (Okun 1970; 
Prachowny 1993), where the increased demand for labour is absorbed internally. This 
is – for the most part – the case of nations, though one might argue that rather small 
countries like San Marino or Luxembourg rely on the open nature of their economies. 
The assumptions of fixed endowment of production factors – one being labour – and 
of constant returns of scale, which are key assumptions of conventional trade models 
such as Heckscher-Ohlin’s (Ohlin 1933), are increasingly undermined by the 
abatement of capital and labour mobility barriers (we can think of free-trade areas and 
the Schengen agreement). Such assumptions also seem to be restrictive when 
considering small open systems such as regions, where an increased demand for 
labour in a particular region may not be sufficiently satisfied internally and is 
complemented by intensified labour mobility (that is, incoming commuting from the 
neighbouring areas). As a consequence, the difference between closed and open 
systems – in particular with regard to regional labour markets – deserves further 
investigation for its implications on the Okun framework described in the previous 
section. 
(b) A second reason for interest in regional trade is related to regional convergence (Barro 
and Sala-i-Martin 1991, 1992). The concept of convergence refers to the economic 
process by which poorer regions tend to catch up over time – for example in per capita 
income – with the richer ones. Though the data employed in the present study do not 
8 Chapter 1 
 
allow us to verify long-term adjustments and trends (see Chapter 3), it is still possible 
to investigate medium-range (5–10 years) tendencies to regional convergence. In this 
regard, the investigation of convergence in the economic output of regions would 
require knowledge of the interregional exchange of goods and services (flows of 
resources). In particular, because such data, as stated above, are not available at the 
required level of detail, it is not possible to shed light on the network of economic 
interrelations that would favour convergence between regions. An alternative approach 
is therefore necessary. 
 
In light of the above discussion, we can again exploit and reinterpret the framework 
provided by Okun’s law. In the preceding section we argued that labour market aggregates, 
such as unemployment rates, provide information on where the economy stands. Likewise, we 
now argue that the daily flows of workers across regions (interregional commuting) are a 
proxy for (a direct and indirect consequence of) the flows of economic resources between 
regions. Following our speculation, regions with a prevailing outward movement of resources 
will show high rates of incoming workers from the neighbouring regions. Accordingly, 
regions which mainly ‘import’ resources will be providers of labour force.
2
 In other words, 
the degree of interdependence among regions is expected to be determined by commuting 
flows.  
The significance of labour mobility (often comprising migration phenomena) as an 
indicator of interregional flows, in particular as an implication of open systems with regard to 
regional convergence, is discussed in the recent literature. Magrini (2004) discusses 
interregional interactions in terms of both trade and labour mobility. In an analysis of the 
Chicago metropolitan area, Hewings et al. (2001) come to a conclusion which also seems to 
support the above approach. The authors find that journey-to-work flows, rather than regional 
trade, generate a closer interdependence among regions. The level of interdependence among 
regions is crucial in determining the extent and duration of differentials among them (in terms 
of economic performance). As Arbia et al. (2002, p. 25) stress, in the presence of strong 
regional interdependence, ‘a region experiencing growth propagates positive effects onto the 
neighbouring regions thus producing an acceleration of the convergence process’. Further, on 
the basis of their experiments concerning Italian regions, the authors find that, by taking into 
consideration – by means of spatial econometric techniques – the levels of regional 
interdependence, a significantly higher degree of convergence is found. 
If the extent of the aforementioned regional interdependencies may be an indicator of the 
interaction externalities, the distribution and persistence over time of these interregional 
liaisons are also important, since they show the direction towards convergence or divergence. 
In view of the commuting flows analogy described above, the distribution of labour mobility 
                                                 
2
  The existence of more specific phenomena such as the ‘dormitory districts’ can instead be attributed to 
distortions caused by zoning/land-use policies, real estate prices and so on. 
 Introduction: Theoretical and Policy Aspects 9 
 
can tell rather different stories: Do regions which are already economically advanced tend to 
receive increasing or decreasing shares of mobility over time? Do regions evolve from being 
providers of resources to being attractors of resources? Do mobility patterns which are stable 
over time imply a lack of convergence? 
The evolution of the mobility network caused by economic activity can provide a dynamic 
outlook on this convergence process (or on the lack of it). Its study, together with the analysis 
and interpretation of regional differentials in labour markets, may allow the questions listed 
above to be answered. 
Finally, it should be remarked that, while the theoretical links between labour markets and 
economic output outlined here, as well as those made in the preceding section, are at the basis 
of the motivations for the present study, our aim is not to test or prove the validity of Okun’s 
law or other economic theories. We limit our scope to the statistical description and 
interpretation of the labour market-related variables discussed above. 
 
1.4  Objectives of the Study 
 
The aim of the present study is the statistical analysis of regional labour market developments 
and disparities in Germany. In particular, our analysis is carried out following two distinct – 
but interrelated – research questions. 
The first research question concerns the statistical analysis and forecast of the key 
variables that characterize the functioning of regional labour markets, notably employment 
and unemployment. In particular, we are interested in selecting and applying appropriate 
novel methodological tools in order to take into account the complex spatio-temporal 
relationships among regions. Hence, as well as using conventional analytical tools such as 
spatial econometrics, we pay considerable attention to neural networks (NNs) and spatial 
filtering techniques. The NN approach provides advantages over conventional statistical 
techniques, in that NNs are nonlinear methodologies which are able to autonomously learn – 
from the data – functional relationships between the variables employed. We aim to integrate 
in a forecasting framework – through the internal complex interactions of the NN paradigm – 
the various forces that drive regional labour market developments. A further step is taken with 
the use of spatial filtering techniques. This approach allows us to take explicitly into account 
the spatial dependence among regions and to generate more appropriate estimates for the 
labour market variables studied. 
The second research question is concerned with the spatial mobility associated with 
regional labour market developments addressed in the first empirical task. By studying 
commuting flows between regions and their evolution over time, we aim to integrate evidence 
from conventional spatial interaction and urban-economic approaches with statistical methods 
emanating from recently developed network analysis frameworks, such as those related to 
scale-free networks and the concept of ‘hubs’. The integration of such approaches allows us to 
10 Chapter 1 
 
further investigate patterns of regional disparity. In particular, we explore the emergence and 
the stability of ‘hubs’ of mobility and, consequently, of regional development. In light of our 
findings, a discussion of the mechanisms which drive regional change is offered. 
Together, the experiments related to our two research questions aim to provide novel 
approaches to regional labour market forecasting and modelling. The two-step approach of 
geography-based econometrics and spatial network modelling offers tools for both 
observing/analysing spatial correlation between regional labour markets and interpreting the 
underlying workers’ mobility levels as an indicator of regional interaction. 
 
1.5  Structure of the Study 
 
The present dissertation gathers a number of empirical studies, which provide a progressively 
closer and in-depth analysis of regional disparities in the labour market (see Figure 1.1). In the 
context of the theoretical framework and of the research questions outlined in the previous 
sections, we employ a varied set of methodologies which are described briefly in Chapter 2. 
As a background specific to the case study of this thesis – the German regional labour markets 
– Chapter 3 provides a concise overview of the recent (post-reunification) economic history of 
Germany, as well as of the research carried out with regard to the aforementioned German 
regional disparities, before finally describing in detail the data employed in the study. The 
first three chapters of this study form Part A of the thesis. 
The empirical studies contained in the dissertation can be divided in two main parts. The 
first empirical part (Part B) is concerned with the first research question outlined in the 
previous section, and involves the utilization of novel econometric approaches for forecasting 
and modelling regional labour market changes. We first deal with regional labour market 
forecasts by means of neural networks, while subsequently employing spatial filtering 
techniques in order to accommodate spatial heterogeneity in regional labour market 
aggregates. The second and last empirical part of the thesis (Part C) is concerned with our 
second research question, and deals with the integration of recently developed network 
theories with conventional spatial interaction modelling. 
Part B of the dissertation includes the first four empirical chapters. In the first three of 
these chapters (4–6) we discuss and test the use of neural network (NN) techniques for the 
forecast of variations in regional labour markets. Several factors make NNs a desirable 
approach in this matter, such as: (a) the imbalance emerging from the availability of data 
which are wider horizontally (that is, the geographical disaggregation) than vertically (that is, 
number of observations in time); and (b) the specification constraints and assumptions typical 
of several conventional econometric techniques and further issues such as model 
identification and multicollinearity. NNs offer an alternative approach, since they bypass all 
the above problems, offering a data-driven, unstructured and nonlinear tool. Using German