The employment - wage pattern conundrum: institutions, policies and labour market outcomes

劳动力市场改革包括一系列广泛的措施，这些措施可能对工资增长产生不同的非线性影响。评估改革效果的现有数据集，如欧盟LABREF数据集、经合组织spider数据库或国际劳工组织劳动力市场政策措施清单，只包括非常广泛的劳动力市场改革类别。欧盟统计局的劳动力市场政策统计和经合发组织就业数据库提供了关于劳动力市场机构的高级数据，包括EPL、UB替代率、ALMP支出和登记参加人数。然而，这些数据集没有充分区分不同类型的改革和福利条件措施。我打算采取一种更全面的方法，并计划收集过去二十年来一系列欧洲经济体实施的劳动力市场改革和最低工资政策的国家一级数据。ICTWSS数据集（Visser 2015）包含了关于劳动力市场机构的综合编纂数据，主要限于工资制定机构和社会合作伙伴，这与CB数据的使用有关。MISSOC数据集包含有关失业福利的相关摘要，而比较失业福利条件和制裁数据集（Knotz & Nelson 2018）记录了过去几十年来失业福利条件和制裁的严格程度。文本挖掘和数据清除工具将有助于收集和整理来自MISSOC和国家来源的文本数据（例如通过使用R包tm）。在数据收集之后，我计划按照三种平行的方法来回答研究问题及其子问题。第一种方法打算把重点放在所有欧洲联盟国家，而第二和第三种方法计划把重点放在有充分数据的欧洲联盟国家的一个子集。在国家层面上，我计划观察重大的劳动力市场灵活性改革对总工资增长的影响。使用时变变量，如EPL强度、UB替代率、CB覆盖率和ALMP支出，将允许将回归不连续设计应用于实施重大劳动力市场改革的时间点。这将使我们能够观察到劳动力市场改革前后变量对总工资增长及其分布的影响的任何中断，并发现国家之间和不同类型改革之间的异质性2。作为补充，我设想对选定的国家采取更细致的方法。我将使用可用国家的微观面板数据（例如德国的SOEP，英国的BHPS/ASHE）和其他国家的重复横截面微观数据（例如欧盟LFS加权的欧盟- silc）来估计劳动力市场灵活性改革对不同收入十分位数的工资增长的影响。研究策略将遵循差异中差异设计，观察菲利普斯曲线的联系和地区、部门、职业和收入十分位数之间的分布差异。在第三种方法中，我计划研究不同政策变化随时间的相互作用。现有的研究方法往往难以研究改革的不同组合及其时间顺序。应用监督机器学习（例如使用R包插入符号）允许研究嵌套数据，这一点尤为重要，因为我们可以预期劳动力市场政策变化的影响依赖于先前改革的路径，并对工资增长及其分布产生非线性影响。

Labour market reforms encompass a wide set of measures, which may have varying and non-linear effects on wage growth. Established datasets to evaluate reform effects such as the EU LABREF dataset, the OECD SPIDERdatabase or the ILO Inventory of Labour Market Policy Measures only include very broad categories of labour market reforms. The Eurostat labour market policy statistics and the OECD employment database provide high-leveldata on labour market institutions including EPL, UB replacement rates, ALMP spending and the number of enrolled participants. However, these datasets do not contain sufficient differentiation between different typesof reforms and measures of benefit conditionality.I intend to take a more comprehensive approach and plan to assemble country-level data of implemented labour market reforms and ALMP policies over the last two decades in a set of European economies. The ICTWSS dataset(Visser 2015) comprises a comprehensive collection of codified data on labour market institutions, mainly limited to wage setting institutions and social partners, which is relevant to use for CB data. The MISSOC dataset containsa relevant summary on unemployment benefits and the Comparative Unemployment Benefit Conditions & Sanctions dataset (Knotz & Nelson 2018) documents the increase in the strictness of unemployment benefitconditions and sanctions over the past decades. Text mining and data scrapping tools will help to collect and codify textual data from MISSOC and national sources (e.g. by using the R package tm). After the data collection, I planto follow three parallel approaches to answer the research question and its sub-questions. The first approach is intended to focus on all European Union countries, while the second and third approaches are planned to focuson a subset of European Union countries for which adequate data are available.1. On a country level, I plan to observe how major labour market flexibilizing reforms play out on aggregate wage growth. Using time-varying variables such as EPL strength, UB replacement rate, CB coverage rateand ALMP spending will allow applying a regression discontinuity design to the point in time when a major labour market reform was implemented. This will allow to observe any break in the effect of the variableson aggregate wage growth and its distribution before and after labour market reforms, and to detect heterogeneity between countries and different types of reforms2. Complementarily, I am envisaging to take a more granular approach for selected countries. I will estimate the effects of labour market flexibilizing reforms on wage growth of different income deciles using micropanel data for available countries (e.g. SOEP for Germany, BHPS/ASHE for the UK) and repeated cross-sectional micro data for others (e.g. EU-SILC weighted by EU LFS). The research strategy will follow adifference-in-difference design to observe the Phillips curve link and distributional differences across regions, sectors, occupations and income deciles.3. In the third approach, I plan to study the interactions of different policy changes over time. Established research methods often struggle to study different combinations of reforms and their time sequencing.Applying supervised machine learning (e.g. using the R package caret) allows studying nested data, which is particularly important as we can expect the effects of labour market policy changes to be path dependenton previous reforms and to have non-linear effects on wage growth and its distribution.