Abstract This paper considers a model that accommodates both output and input-specific inefficiency components (input slacks). We use a translog function to represent the underlying production technology in which the input slacks are generalized to have both deterministic (functions of exogenous variables) and stochastic components. Consequently, the composed error term becomes a nonlinear function of several error components, viz., a one-sided input slack vector (the dimension of which depends on the number of inputs), a one-sided output technical inefficiency and a two-sided random noise. Identification of two sets of one-sided errors is possible in a translog model because the vector of one-sided input slacks appears in additive form as well as interactively with the (log) inputs. Distributional assumptions on technical inefficiency and slacks also help in identification. Bayesian inference techniques are introduced, organized around Markov Chain Monte Carlo, especially the Gibbs sampler with data augmentation, to estimate these inefficiency components. For an empirical application we use a large unbalanced panel of the U.K. manufacturing firms. Slacks associated with labor and capital are found to be 2.35% and 10.74%, on average. Output (revenue) loss from technical inefficiency is, on average, 2.43%, while revenue loss from input slacks is, on average, 9.2%.

中文翻译：

---

投入产出效率剖析：广义随机前沿模型

摘要 本文考虑了一个模型，该模型同时包含输出和特定于输入的低效率组件（输入松弛）。我们使用 translog 函数来表示底层生产技术，其中输入松弛被概括为具有确定性（外生变量的函数）和随机分量。因此，组合误差项变成了几个误差分量的非线性函数，即单边输入松弛向量（其维度取决于输入的数量）、单边输出技术低效率和两侧随机噪声。在 translog 模型中可以识别两组单边错误，因为单边输入松弛的向量以加法形式出现，并且与（对数）输入交互。关于技术效率低下和松弛的分布假设也有助于识别。贝叶斯推理技术被引入，围绕马尔可夫链蒙特卡罗组织，特别是具有数据增强的吉布斯采样器，以估计这些低效率组件。对于实证应用，我们使用英国制造公司的大型不平衡面板。与劳动力和资本相关的松弛平均为 2.35% 和 10.74%。技术效率低下造成的产出（收入）损失平均为 2.43%，而投入不足造成的收入损失平均为 9.2%。对于实证应用，我们使用英国制造公司的大型不平衡面板。与劳动力和资本相关的松弛平均为 2.35% 和 10.74%。技术效率低下造成的产出（收入）损失平均为 2.43%，而投入不足造成的收入损失平均为 9.2%。对于实证应用，我们使用英国制造公司的大型不平衡面板。与劳动力和资本相关的松弛平均为 2.35% 和 10.74%。技术效率低下造成的产出（收入）损失平均为 2.43%，而投入不足造成的收入损失平均为 9.2%。