Using deep Q-learning to understand the tax evasion behavior of risk-averse firms
作者：
Highlights：
• A Markov-based decision support model with non-linear reward function.
• Complexity is tackled computationally via Deep Q-learning.
• We evaluate various tax policy scenarios including occasional tax amnesties.
• We compute a firm’s “average” risk-aversion coefficient from empirical data.
• Tax-amnesties have a negative long-term impact on tax revenues.
摘要
•A Markov-based decision support model with non-linear reward function.•Complexity is tackled computationally via Deep Q-learning.•We evaluate various tax policy scenarios including occasional tax amnesties.•We compute a firm’s “average” risk-aversion coefficient from empirical data.•Tax-amnesties have a negative long-term impact on tax revenues.
论文关键词：Markov decision processes,Tax evasion,Q-learning,Deep learning
论文评审过程：Received 27 October 2017, Revised 9 January 2018, Accepted 25 January 2018, Available online 1 February 2018, Version of Record 27 February 2018.
论文官网地址：https://doi.org/10.1016/j.eswa.2018.01.039
