Some Properties of a Modification of the Limited Information Estimator

What drives international financial flows? Politics, institutions and other determinants

“…Third, to tackle endogeneity and account for measurement error in the institutional proxies, I estimate instrumental variable models using legal origin (following La Porta et al 1998) and population density before colonization (following Acemoglu et al 2002) to extract the historically predetermined component of institutions on foreign bank flows. The IV estimates imply a highly significant and economically sizable effect of institutional perfor- 25 In unreported models I also followed the approach of and distinguished between property rights and contractual institutions using their preferred measures. I found that both institutional types are significant correlates of bank flows.…”

“…Fuller LIML with a constant of 1 is the best unbiased estimator, while when the constant is 4 the procedure minimizes the mean squared error of the estimator (Fuller, 1977). The estimates with the Fuller LIML methods are somewhat smaller to the corresponding 2SLS coefficients, but more precisely estimated.…”

“…The results imply that the key impediment to foreign bank lending is a high risk of expropriation from the government and a significant likelihood that due to legal system inefficiencies the borrower will repudiate the international loan contract. 25 …”

The economic costs of court decisions concerning dismissals in Japan: Identification by judge transfers

“…LIML is proved to possess better finite sample properties than 2SLS, while LIML and 2SLS share the same asymptotic distribution. On the other hand, LIML is also known to have no moments, and Fuller's (1977) In fact, partial R-squared here equals 0.363, and judges effects appear to be relevant rather than weak instruments in the first stage regression. This supports observations in section 3.2 that judges do matter in trials in Japan.…”

“…According to the critical values presented in Stock and Yogo (2004), this suggests that the bias of 2SLS may be 10 to 20% that of OLS. Thus, the estimates with three other alternative methods are also presented in order to mitigate the finite sample bias problem-Limited Information Maximum Likelihood (LIML), Fuller's (1977) modification of LIML (Fuller) 29 , and Jackknife IV estimator proposed by Angrist et al (1999) (JIVE). 30 The first two, as well as 2SLS, are k-class estimators.…”