瑞士国家银行(Swiss National Bank,缩写SNB)瑞士的中央银行,是根据1905年联邦宪法创建的联合股份银行。资本额为5000万瑞士法郎,实收资本2500万瑞士法郎,多数股份由州政府和州银行持有,其余股份由私人持有。1907年在伯尔尼、巴塞尔、日内瓦、苏黎世、圣加伦开始营业。董事会大部人选由联邦政府指派,向联邦议院负责。
This study considers the practically important case of nonparametrically estimating heterogeneous average treatment effects that vary with a limited number of discrete and continuous covariates in a selection-on-observables framework where the number of possible confounders is very large. We propose a two-step estimator for which the first step is estimated by machine learning. We show that this estimator has desirable statistical properties such as consistency, asymptotic normality, and rate double robustness. In particular, we derive the coupled convergence conditions between the nonparametric and the machine-learning steps. We also show that estimating population average treatment effects by averaging the estimated heterogeneous effects is semiparametrically efficient. The resulting estimators are compared to other suggestions in the literature in Monte Carlo experiments that are inspired by real data. They are found to perform relatively better in most settings. The new estimators are applied to the empirical example of the effects of mothers' smoking during pregnancy on the birthweight of their babies.
Migration has long been considered one of the key mechanisms through which labor markets adjust to economic shocks. In this paper, we analyze the migration response of American workers to two of the most important shocks that hit US manufacturing since the late 1990s – Chinese import competition and the introduction of industrial robots. Exploiting plausibly exogenous variation in exposure across US local labor markets over time, we show that robots caused a sizable reduction in population size, while Chinese imports did not. We rationalize these results in two steps. First, we provide evidence that negative employment spillovers outside manufacturing, caused by robots but not by Chinese imports, are an important mechanism for the different migration responses triggered by the two shocks. Next, we present a model where workers are geographically mobile and compete with either machines or foreign labor in the completion of tasks. The model highlights that two key dimensions along which the shocks differ – the cost savings they provide and the degree of complementarity between directly and indirectly exposed industries – can explain their disparate employment effects outside manufacturing and, in turn, the differential migration response.
Double/debiased machine learning (DML) uses for estimating an average treatment effect (ATE) a double-robust score function that relies on the prediction of nuisance functions, such as the propensity score, which is the probability of treatment assignment given covariates. Estimators relying on double-robust score functions are highly sensitive to errors in propensity score predictions. Machine learning algorithms have been found to produce models that often overestimate or underestimate these probabilities. Several calibration approaches have been proposed to improve probabilistic forecasts of machine learners. This paper explores their integration into the DML framework, showing via simulations that using calibrated propensity scores significantly reduces the root mean squared error of ATE estimates in finite samples while preserving DML's asymptotic properties.
This paper examines the impact of Ethiopia's Road Sector Development Program (RSDP) from 1997-2016 on land use and economic activity, using spatial variation in road upgrades and satellite imagery. We use three approaches to triangulate the effect of the RSDP: difference-in-differences examining economic activity near upgraded roads, long-difference instrumental variables comparing areas connected to upgraded versus non-upgraded roads, and market access analysis. For RSDP phases I-III, upgrades increase local economic activity, with effects varying by baseline activity. Areas with medium-to-high baseline activity showed positive effects, while areas with the lowest activity showed minimal effects. Results are weaker for RSDP IV.
Classical neural network approximation results take the form: for every function f and every error tolerance ϵ > 0, one constructs a neural network whose architecture and weights depend on ϵ. This paper introduces a fundamentally different approximation paradigm that reverses this quantifier order. For each target function f, we construct a single recurrent neural network (RNN) with fixed topology and fixed weights that approximates f to within any prescribed tolerance ϵ > 0 when run for sufficiently many time steps.The key mechanism enabling this quantifier reversal is temporal computation combined with weight sharing: rather than increasing network depth, the approximation error is reduced solely by running the RNN longer. This yields exponentially decaying approximation error as a function of runtime while requiring storage of only a small, fixed set of weights. Such architectures are appealing for hardware implementations where memory is scarce and runtime is comparatively inexpensive.To initiate the systematic development of this novel approximation paradigm, we focus on univariate polynomials. Our RNN constructions emulate the structural calculus underlying deep feed-forward ReLU network approximation theory—parallelization, linear combinations, affine transformations, and, most importantly, a clocked mechanism that realizes function composition within a single recurrent architecture. The resulting RNNs have size independent of the error tolerance ϵ and hidden-state dimension linear in the degree of the polynomial.