Modelling and Simulating the Effect of Wages on East-West German Migration Patterns since 1989
Frank Heiland, State University of New York at Stony Brook
1. Objectives
The transformation of the Eastern European Economies since the late 80th has induced the EC economic commission to put International Migration on the top of its policy agenda. Many important demographic and economic insights can be obtained from the recent East-West migration experience in Germany: From January 1989 to January 1994 about 1,400,000 East Germans, or roughly 8 per cent of the population (15 per cent of the labor force), migrated to West German states. The recent and ongoing debate over the emergence and persistence of high unemployment in East Germany has brought to the forefront fundamental difference of opinion about the rapid adjustment of East German wages to the West German wage level since 1990. At the center of the controversy is the conjecture that a slower wage adjustment would have resulted in significantly more emigration.
In this paper, I examine this hypothesis by analyzing the effect of (expected) income on the decision of East Germans to migrate to the West and the feedback of migration on regional labor market conditions.
The investigation contains two steps:
(1) First, I find good predictors of the decision to migrate from a State in the East to a state in the West in some year from a set of socio-economic variables using multivariate panel regression technique and state-level aggregate data.
(2) Based on the main determinants of the migration decision as found in the multivariate regression analysis (1), I specify a utility maximization problem using a finite horizon, discrete choice dynamic programming model. Each period a representative agent located in an East German state chooses whether to stay or to migrate to one of the eleven West German states. The value of each choice is effected by the current employment status which itself depends in part on choices made in the past. Moreover, the aggregate migration outcome effects state-level unemployment rates.
2. Background
Previous work on East-West German migration has looked at the determinants of the migration decision (Burda (1993), Burda et. al. (1998), Hunt (2000)) using both micro-level and state-level data from mid respectively late 1990. The descriptive analysis in my work extends this literature by including data from 1989.
To my knowledge there has been no previous attempt to model East-West German migration using a discrete dynamic choice model. Since the migration decision is clearly state-contingent and involves forward-looking behavior, accounting for the dynamic nature of the migration problem is an important generalization that can provide better insights in the decision making process. Furthermore, this research is first to address the interaction between the migration outcome and regional labor market conditions in the context of East-West German migration. This allowance for macroeconomic feedbacks is necessary to obtain an unbiased effect of wage adjustment on migration propensities.
3. Data and Methods
The multivariate panel-regression analysis is carried out based on state-level-aggregate data for all German states except for Brandenburg from the Department of Residency of the German Democratic Republic (ZER, Berlin-Biesdorf), and from the Department of Statistics of the Federal Republic of Germany as reported in various German Statistical Yearbooks. The statistical model with fixed effects in the departure and random effects in the destination states is estimated using feasible generalized least squares (see for example Greene (2000)). A Breusch and Pagan Lagrange Multiplier test and a Hausman test for random effects are performed.
Since there is no ready-built software to solve Dynamic Discrete Choice Models I program the sequential decision making process using the FORTRAN90 language. The code solves the finite time utility maximization problem using a recursive algorithm that yields the state contingent optimal decision rule for each potential migrant. A fixed-point algorithm guarantees that in the optimum people anticipate future state-level unemployment rates correctly (rational expectation equilibrium). Conditional on a set of parameters such as subjective time preference rate, and travel cost and a relative wage adjustment path, the solution of the model is given by state-level migration propensities and unemployment rates.
4. Achieved Research Tasks
(1) Multivariate Regression Analysis of State-Level Migration Propensity, 1989-1994
Including data for 1989 in the regression analysis yields results consistent with the previous literature (see 4./5.): There is compelling evidence that migrants are more attracted to Western States that are relatively economically sounder. The propensity to move to a West German state increases as the level of the unemployment rate decreases in that state. The higher the relative earnings from migrating to a state, the greater is the chance of moving there. The return to human capital in the destination state is one of the most important determinants of East West migration: a target state with a 10 per cent higher human capital index attracts (ceteris paribus) almost 2 additional migrants per thousands. As expected, cost factors are negatively associated with the chance of inflow. The higher the price of residential land is in the destination state relative to the home state the smaller the stream of migrants in that direction. Furthermore, the distance between the states is crucial to the migration decision: the propensity to migrate to a state decreases by almost 1 per thousands of the home state's population for each additional 100 kilometers of distance between an East and a West German state. Finally, the estimated coefficient of (agri)cultural difference suggests that the more the target state's economy differs with regard to its share of agricultural in production the smaller the chance that people from this region in the East will move to that state in the West.
(2) Discrete Dynamic Choice Simulation Model of Migration with Macro-Feedback
The code to solve the decision problem with macroeconomic feedback is written. Preliminary simulation runs suggest that the model is capable of reproducing the observed patterns of the migration propensities over time and across states by using a small number of parameters: travel cost and cultural similarity coefficient in the potential migrant's problem and labor market response parameters at the macro-level.
A preliminary wage-adjustment policy experiment suggests that a decline in the relative wage rate would have been associated with a significant increase in emigration rates and lower unemployment rates in the East German states.
5. Current and Future Work
Current work includes efforts to improve the calibration of the simulation model. Next, I plan to analyze how the following policies effect the migration flows and the unemployment rates:
· unemployment support,
· wage differential, and
· travel cost subsidies.
Future work contains additional degrees of heterogeneity among the agents. Based on the empirical findings referenced previously, most promising dimensions of observable variation at the individual level are:
· age,
· skills level, and
· unemployment duration.