We propose a novel method for modelling income processes using machine learning. Our method links age-specific regression trees, and returns a discrete state process, which can easily be included in consumption-saving models without further discretizations. A central advantage of our approach is that it does not rely on any parametric assumptions, and because we build on existing machine learning tools it is furthermore easy to apply in practice. Using a 30-year panel of Danish males, we document rich higher-order income dynamics, including substantial skewness and high kurtosis of income levels and growth rates. We also find important changes in income risk over the life-cycle and the income distribution. Our estimated process matches these dynamics closely. Using a consumption-saving model, the implied welfare cost of income risk is more than 10% of income.

中文翻译：

---

链接回归树的高阶收入动态

我们提出了一种使用机器学习对收入过程进行建模的新颖方法。我们的方法链接特定年龄的回归树，并返回离散状态过程，该过程可以轻松地包含在节省能耗的模型中，而无需进一步离散化。我们方法的主要优势在于它不依赖任何参数假设，而且由于我们基于现有的机器学习工具，因此在实践中更容易应用。我们使用由30年的丹麦男性组成的小组，记录了丰富的高阶收入动态，包括收入水平和增长率的严重偏度和高度峰度。我们还发现，在生命周期和收入分配中，收入风险发生了重要变化。我们估计的过程与这些动态紧密匹配。使用节省费用的模型，