Ownership and Financial Sustainability of German Acute Care Hospitals (original) (raw)

Econometrics Journal (2008), volume 11, pp. 349–376. doi: 10.1111/j.1368-423X.2008.00242.x

2016

Summary We present a generalized LM test of heteroscedasticity allowing the presence of data transformation and a generalized LM test of functional form allowing the presence of heteroscedasticity. Both generalizations are meaningful as non-normality and heteroscedasticity are common in economic data. A joint test of functional form and heteroscedasticity is also given. These tests are further ‘studentized ’ to account for possible excess skewness and kurtosis of the errors in the model. All tests are easy to implement. They are based on the expected information and are shown to possess excellent finite sample properties. Several related tests are also discussed and their finite sample performances assessed. We found that our newly proposed tests significantly outperform the others, in particular in the cases where the errors are non-normal.

A Small-Sample Correction for Testing for gth-Order Serial Correlation with Artificial Regressions

Boston College Working Papers in Economics, 1996

Monte Carlo experiments establish that the usual \t-statistic" used for testing for rst-order serial correlation with articial regressions is far from being distributed as a Student's t in small samples. Rather, it is badly biased in both mean and variance and results in grossly misleading tests of hypotheses when treated as a Student's t. Simply computed corrections for the mean and variance are derived, however, which are shown to lead to a transformed statistic producing acceptable tests. The test procedure is detailed and exemplar code provided. whether the OLS estimate c of the coecient of the lagged residuals in this regression is signicantly dierent from zero. Asymptotically, a t-test based on a Student's t distribution with appropriate degrees of freedom is relevant for this purpose, but, of course, interest centers here on whether this remains so for truly small samples. As a practical matter, it makes no dierence whether e n or y n appears on the left-hand side of this articial regression. Since e n y n x T n b OLS , substituting y n for e n would result in dierent estimates for , but would have no eect on c or its variance, and hence on the value of t c , the standard t-statistic for c. This latter statistic is derived formally in Appendix C, but in essence it is determined as follows: Let X [g] be the (N g)K matrix that results from deleting the rst g rows from X, and let e g be the (N g)-vector obtained by lagging the residual vector e y Xb OLS g times (i.e., removing its last g elements). Dene the (N g)(K +1) matrix Z [X [g] e g ], and let s 2 c denote the (K +1)st diagonal element of s 2 (Z T Z) 1 , where s 2 v Tv =(N K g 1),v being the residuals from the articial regression of y [g] (y with its rst g rows deleted) on Z, i.e., y [g] X [g] d + ce g + v. Then t c = c s c : (2:6) 3. The non-Studentness of t c in small samples Monte Carlo experiments are used to demonstrate the divergence of the small-sample distribution of t c from a Student's t. A Student's t with r degrees of freedom, we recall, is a symmetric distribution with mean zero and variance r=(r 2). Relative to this, we nd for small samples that t c is biased (often severely), has depressed variance, and is asymmetric. Monte Carlo experiments for this use are quite straightforward. Twenty-two models, diering in the structure of the data comprising X, are examined for N = 20, a very

A joint test for serial correlation and heteroscedasticity

Economics Letters, 1984

The problem of testing Jointly for first-order autoregressive and heteroscedastic disturbances in the linear regression model is considered. A test which is most powerful invariant in a predetermined neighbourhood of the alternative hypothesis parameter space is proposed.

Who Gets the Credit? Determinants of the Probability of Default in the German Hospital Sector

SSRN Electronic Journal, 2006

Huge underinvestment increases the need for private borrowing in the German hospital sector, the access to which is partly determined by the probability of default (PD) of individual hospitals. Using ordinary least squares and quantile regression techniques this paper provides first empirical evidence of its kind to evaluate the PD in the hospital sector and its constituent determinants. Based on annual account and medical data from 17% of all German hospitals we find that the current average probability of default amounts to approximately 1.7%, which is slightly higher than the average probability for all German firms. Among other determinants, we find that public ownership significantly increases the risk of default, while private for-profit and private not-for-profit hospitals do not differ. Moreover, demographic change in the form of population growth is confirmed to be relevant for the PD.