Sascha G. Wolf, Model Specification II: Error Correction Model in:

Sascha G. Wolf

Pharmaceutical Expenditure in Germany, page 60 - 64

Future Development, Political Influence and Economic Impact

1. Edition 2009, ISBN print: 978-3-8329-4164-2, ISBN online: 978-3-8452-2005-5

Series: Neue Studien zur Politischen Ökonomie, vol. 6

Bibliographic information
60 Model 5, which combines opportunistic, partisan and corporatistic behaviour, delivers the best results not only in respect to the adjusted, but also regarding the signi? cance of the exogenous variables. To recapitulate, we found an indication for partisan politics as well as for the combination of partisan, opportunistic and corporatistic behaviour. Furthermore, the age structure of the population and co-payments of the insurants obviously in? uence drug expenditure, whereas the effects of income on expenditure development are ambiguous. 3.5.5. Model Speci? cation II: Error Correction Model (ECM) Taking ? rst differences to correct for non-stationarity leads to a loss of valuable information, especially where a long-run trend in the variables is concerned. To overcome this problem and as an additional test of signi? cance, we also estimated an ECM. Beforehand, examining whether the variables are co-integrated, i.e. that there exists at least one linear combination that is integrated of order zero (I(0)), is recommended.47 If the variables are co-integrated, then the non-stationarity in the variables cancel each other out and it is possible to apply an ECM. Since we have more than two variables, we use Johansen’s method to test for cointegration (Johansen 1988). Table 3.3: Johansen Cointegration Test. Maximum Eigenvalue Statistics Trace Statistics None 36,521 (27,584)* 60,53204 (47,856)** At most 1 15,701 (21,132) 24,01055 (29,797) At most 2 6,827 (14,265) 8,309441 (15,495) At most 3 1,483 (3,841) 1,482590 (3,841) Note: The values in parentheses show the 5% critical value. Trend assumption: linear deterministic trend (unrestricted). *Maximum-Eigenvalue test indicates 1 cointegrating equation at the 5% level. **Trace test indicates 1 cointegrating equation at the 5% level. Both the maximum eigenvalue and the trace test indicate that there exists a single cointegration relationship.48 Thus we should apply the Engle-Granger two-step method for specifying an ECM.49 47 For more on cointegration, see Engle and Granger (1988). 48 Since the Johansen Cointegration Test indicates a single cointegration relationship, the residuals of the OLS regression have also been tested for a unit root (Engle and Granger 1987). The Augmented Dickey-Fuller as well as the Phillips-Perron test con? rms the existence of a cointegration relationship. 49 Whenever there exists a single cointegration relationship, the Engle-Granger two-step method delivers more robust results than the Johansen one-step method (Kennedy 2003, p. 340). 61 In the ? rst step, the cointegration relationship which re? ects long-run equilibrium must be estimated. We test the following cointegration relationship: (3.3) . Assuming that the age structure of the population and the wages of SHI members are following a trend, a time trend variable (year) has been included. Additionally, it can be supposed that drug expenditure are also affected by trends due to technological progress which is not totally represented by the wage development. Including a time trend means that if the system were always in equilibrium, then the variables would grow over time (Kennedy 2003, p. 339). In the second step, we formulate an ECM by extending the ? rst-differenced model (section 3.5.3) with a constant and the lagged residuals (?) of the cointegration relationship: (3.4) . The lagged residuals of the cointegration relationship (?t-1) represent the error correction term in the ECM. The coef? cient is expected to have a negative sign. The error correction term balances deviations from the long-run equilibrium. In contrast, the ? rst-difference term shows short-run relationships.50 In total, the results of the ECM (table 3.4) prove the outcomes of the afore estimated ? rst-differenced regressions in section 3.5.4. Once again, the best results deliver the purely partisan model 3 and the combined behaviour model 5. In both regressions, the dummy variable is signi? cant and the adjusted R2 shows a very good ? t having a value of 75 and 72 % respectively. Age structure as well as co-payments are highly signi? cant on a 1 % level; all other estimations are inferior. Once again the purely opportunistic model 2 reveals, with a critical low Durbin/Watson statistic and nonsigni? cance of age structure, an arguable speci? cation. Model 4 improves the results, but stays below models 2 and 5. The error correction term is clearly different from zero, indicating again that the variables are co-integrated and hence using an ECM is possible. 50 For further information about ECM see Engle and Granger (1987). 62 Table 3.4: Regression Results: Error Correction Model (ECM). Variable Model 1 (Basic) Model 2 (Opportunistic) Model 3 (Partisan) Model 4 (Partisan & Opportunistic) Model 5 (Partisan & Opportunistic & Corporatistic) Constant term 0,021* (1,759) 0,023* (1,933) 0,015 (1,415) 0,013 (1,021) 0,009 (0,644) Age60 1,043** (2,261) 0,828 (1,657) 1,241*** (2,971) 1,399** (2,698) 1,513*** (2,976) BWage 0,329 (0,823) 0,069 (0,148) 0,131 (0,36) 0,516 (1,252) 0,623 (1,517) CPay -0,146*** (-4,798) -0,141*** (-4,621) -0,161*** (-5,807) -0,159*** (-5,11) -0,163*** (-5,404) Error Correction term -0,928*** (-3,241) -0,861*** (-2,956) -0,806*** (-3,111) -0,92*** (-3,297) -0,915*** (-3,397) PDum -0,02 (-1,087) 0,017** (2,322) 0,01 (1,374) 0,013* (1,769) Adj. R2 0,681896 0,685465 0,750411 0,698607 0,71924 Durbin/Watson 1,579 1,533 1,764 1,88 2,044 Note: *** indicates signi? cance at the 1% level, ** at the 5% level, and * at the 10% level. At ? rst glance the persistent non-signi? cance of wages is surprising, but this can be explained quite easily. The ECM represents only short-term effects. The long-run equilibrium, demonstrated by the cointegration relationship (table 3.5), indeed ? nds high signi? cance for the total base wage of the SHI members. Table 3.5: Regression Results: Cointegration Relationship. Variable Coeffi cient estimate (Standard error) Constant Term -2,44 (-1,414) Age60 1,148*** (7,703) BWage 1,006*** (5,727) CPay -0,133*** (-3,996) Trend 0,051* (2,09) Adj. R2 0,985708 Durbin/Watson 1,662 Note: *** indicates signi? cance at the 1% level, ** at the 5% level, and * at the 10% level. 63 Consequently, in the long run income development is connected to pharmaceutical expenditure, which shows that higher society prosperity leads to higher health care expenditure.51 Thus, in the long-term the richer a society is, the more comprehensive medical provisions are due to higher quality and improved medical treatments, but drug consumption does not vary with short-term changes in income. This is due to the data-set used, which predominantly considers prescribed drugs. In Germany, the requirement for a prescription depends on the degree of adverse reactions of a pharmaceutical product. Thus, in contrast to drugs for the heightening of general well-being, few incentives exist to consume drugs which are reimbursed, i.e. drugs with adverse reactions, without being ill. People consume prescribed drugs primarily in cases of a health disturbance and not because of higher incomes. Furthermore physicians, not patients, decide on medical therapy. It can be presumed that physicians reach their prescription decisions independently from the ? nancial situation of their patients. Taken together, the results con? rm the hypothesis that governments use health policy for achieving sel? sh intentions, especially to serve the interests of their clientele. According to partisan theory, right-wing governments show no cost-cutting ambitions at the expense of the suppliers, but encourage increasing drug expenditure to strengthen the international position of the pharmaceutical industry and for the economic prosperity of the health care suppliers, respectively. In contrast, left-wing governments try to disburden the SHI insurants at the expense of the industry to avoid ? nancial overloads of their electorate, who largely belong to the low-income social classes. Additionally, evidence can be found for the combination of partisan, opportunistic and corporatistic behaviour. During a Christian-Liberal coalition, before elections votemaximizing strategy dominates and leads to temporary cost-containment. This behaviour complies with opportunistic theory. For left-wing governments, opportunistic behaviour cannot explicitly be identi? ed because partisan and opportunistic politics act in the same direction. Our approach will be completed by considering the interests of companies in the pharmaceutical business. Industry has no ambition to weaken the re-election chances of the right-wing coalition and accepts decreasing drug expenditure before elections in the anticipation of re-increasing drug turnovers afterwards. Since the government change in 1998, the political position of the manufacturers has declined. Actually, our regression results give rise to the supposition that the suppliers have consciously counteracted the re-election chances of the Red-Green government; at a minimum they have no ambition to support the left-wing coalition. 51 This result coincides to the ? ndings of Breyer and Ulrich (2000) for total health care expenditure. 64 3.6. Empirical Observations Unexpected regularities in the development of pharmaceutical expenditure can be found (? gure 3.7): During the era of the Christian-Liberal coalition, which lasted until 1998, after two to three years of growth, one year with comparably drastically low or even negative growth rates followed. Thus, the health care reform acts of 1989 (GRG), 1993 (GSG) and 1997 (1st and 2nd SHI Restructuring Act) did indeed have cost-cutting short term in? uences on the expenditure growth rates. But this recovery always remained only in the year of the introduction of the reform act and afterwards it disappeared again in support of a new cost increase. This observation allows three conclusions: ? rstly, government is able to affect drug expenditure. Secondly, the cost-driving protagonists of the health care system need only some months to ? nd new ways of strategic behaviour to circumvent the recently established control instruments; and thirdly, the legislature was either not able or not willing to intervene into the market every year, but rather in important years. Considering the elections for the German Bundestag (the black pillars in ? gure 3.7), we can ? nd a remarkable clear pattern: During the Christian-Liberal coalition, the local minima of the growth rates are always to be found in the year before an election. We can adhere to the statement that apparently the right-wing government tried to cut drug expenditure down before elections, but afterwards it no longer had any reasons to prevent increasing pharmaceutical costs. This observation leads us to the presumption that it appears to be attractive for politicians to cut the costs at the expense of the pharmaceutical industry to gain votes. After the elections, the interests of the providers attain political prevalence over the cost-containment aims of the government and the growth rates increase again. Figure 3.7: Pharmaceutical Expenditure Growth Rates and Elections for the German Bundestag. -13 -8 -3 2 7 12 19 83 19 87 199 1 199 5 199 9 200 3 Growth Rates in % Year Source: Based on data from the Federal Statistical Of? ce and Schwabe and Paffrath (several volumes). From 1992 this includes the new German Laender.

Chapter Preview



Der Arzneimittelsektor der Gesetzlichen Krankenversicherung stand wiederholt im Fokus zahlreicher Gesundheitsreformen. Dennoch ist es bislang nicht gelungen, den Trend steigender Ausgaben nachhaltig zu bremsen. Die vorliegende Untersuchung leistet einen Beitrag dazu, die Ursachen dieser Entwicklung zu erklären und Lösungsansätze aufzuzeigen. Mittels Hauptkomponenten- und Cluster-Analyse wurden Gruppen von Arzneimitteln mit vergleichbaren Konsumeigenschaften gebildet. Jede Gruppe wurde auf den Einfluss der Altersabhängigkeit und des technologischen Fortschritts hin analysiert. Aufbauend auf diesen Ergebnissen wurde eine Prognose der zukünftigen Ausgabenentwicklung bis zum Jahr 2050 erstellt. Obwohl die Hauptkostenfaktoren exogen sind, steht der Gesetzgeber dem vorhergesagten ansteigenden Kostenpfad nicht hilflos gegenüber. Im Gegenteil: Anhand ökonometrischer Tests wird gezeigt, dass die Gesundheitspolitik in der Vergangenheit durch wahl- und klientelorientierte Interessendurchsetzung geprägt war. Mehr Effizienz in der Arzneimittelversorgung könnte durch die Einführung individueller Gesundheitssparkonten erzielt werden. Dies bestätigen die Resultate eines vertikal differenzierten Wettbewerbsmodells.