Thomas Piketty是谁

Dracula

晨枫 发表于 2014-4-23 21:31
多谢资料。不过这里还是语焉不详啊。比如说，个人收入中，到底是精确到实际收入，还是income bracket？估 ...

经济学经验研究中使用的数据要比自然科学质量低不少。关于收入分配的研究也不例外。Piketty也很清楚这一点，在他的论文和新出的这本书里反复提到。不过，在他做政策建议，主张提高对富人的税收，增加再分配的时候，似乎就忘了他所依赖的数据不是那么可靠。

你如果对他的研究和收集数据的方法感兴趣的话，可以读一下他的论文。

Income Inequality and Progressive Income Taxation in China and India, 1986-2015

http://piketty.pse.ens.fr/fichie ... yQian2009_AEJPP.pdf

这是其中关于中国数据来源的部分。

The Chinese data used in this paper comes from the urban household income surveys collected by China's National Statistical Bureau (NSB). These surveys are designed so as to representative of urban China. Between 13,000 and 17,000 households are surveyed each year until 2002, where the sample rose to 45,000-50,000 in 2002 and 2003. The micro-files for these surveys are unfortunately not available for all years, and we asked the NSB to provide us with annual, 1986-2003 tabulations based on the micro-files. We asked for two series of tabulations: household tabulations and individual tabulations. Household tabulations report the number of households whose total household income falls into that bracket, their average total income and household size, as well as their average income broken down by income sources (wage income, business income, capital income and transfer income) for a large number of income brackets (and in particular a large number of top income brackets). Individual tabulations report the number of individuals whose individual income falls into that bracket, their average age, years of education, income and household size, as well as their average income broken down by income sources for a large number of income brackets. In practice, some forms of income cannot be properly attributed to a specific individual within the household (this is particularly true for transfer income and capital income). Hence, the total income aggregates reported in household tabulations are larger than in individual tabulations, and various adjustments are necessary when one uses the latter. However, the  important advantage of individual tabulations is that China's income tax applies to individual income (rather than household income).

We used standard Pareto interpolation techniques to approximate the form of the Chinese household and individual distribution of income, and we then used these structural parameters to compute top fractiles’ incomes and to make income tax simulations. The Chinese data appears to be very well approximated by a Pareto distribution (for any given year, Pareto coefficients are extremely stable within the top decile), although there is some presumption that top incomes are underestimated in the survey data. For each year of the 1986-2003 period, we computed income thresholds and average incomes for fractiles P0-90, P90-95, P95-99, P99-99.5, P99.5-99.9 and P99.9-100. Projections for the 2004-
2015 period were made by assuming nominal income trends by fractile similar to those observed during the 1996-2003 period.

We did not attempt to use similar tabulations from rural household surveys. According to the 2000 China Population Census, over 97 percent of households in rural areas are agricultural households, and are hence exempt from income tax. Average rural income was in 2001 more than 3 times smaller than average urban income. So given that our focus is on top incomes and progressive income taxation, the exclusion of rural households should not be too problematic. In fact, our simulated income tax revenues (based solely on urban household surveys) appear to be reasonably close to actual income tax revenues, both in levels and in trends.

总结一下，他用的数据来自国家统计局1986-2003的问卷调查。调查的问题是收入的bracket。然后假设数据遵循Pareto分布，用bracket的数据来估计这个分布。这种数据的问题自然多多，Piketty对此很清楚。在下一篇我给出链接的文章里，他还讨论过为什么他认为税务局提供的数据优于问卷调查，感兴趣的话，你可以读一下。但是研究中国收入分配的变化毕竟是个非常有意思，非常重要的问题，在没有更好的数据之前，他的研究还是会给我们提供不少有用的信息，另外如果导致Survey数据误差的因素，比如non response，measurement error等等如果不随时间变化的话，那么尽管数据的level不那么可靠，数据的trend 还是比较可靠的。当然这是一个很大的假设，社会经济的变化可能导致误差发生变化，下面的那篇文章里还提到survey的方法变化同样可能导致数据发生大的变化，尽管收入分配变化不大。因此，我觉得在这些数据为基础上的政策建议要谨慎一些。

还有一篇文章，可以说是他在收入分配经验研究工作的综述。

Top Incomes in the Long Run of History

http://piketty.pse.ens.fr/fichie ... ez2011_20110913.pdf

里面还有不少篇幅讨论使用税务局数据估计收入分配的局限性，这些不是我的研究领域，但是读来觉得很有意思。

Dracula

晨枫 发表于 2014-4-24 02:36
"在没有更好的数据之前，他的研究还是会给我们提供不少有用的信息"

完全同意。这样的研究肯定是必要的和 ...

他的研究中计算极富人群的收入，不是直接从数据里来的。而是用bracket的数据先估计一个Pareto分布，然后用这个分布计算出来的。Pareto分布是Pareto100多年前发现这个概率分布描述收入分布非常好，据Piketty说，后来的大量研究都证实了这一点。

Survey里，富人肯定会under-report，就是税务局的数据，也肯定有很多合法非法的偷税漏税，（他还提到税法本身随时间变化也是个问题），survey还有一个问题就是non-response，如果non-response rate富人穷人是一样的话，也不太要紧，如果不一样的话，就会带来很大误差。要纵向比较的话，如果这些导致误差的因素不随时间变化，到也关系不大。不过我对这个假设不是多么有信心。横向各个国家比较的话，就是使用税务局的数据我觉得也不是特别靠谱。象欧洲国家，北欧国家偷税漏税的比例很低，而南欧国家意大利，希腊等各个阶层都对偷税漏税习以为常。Piketty的看法是，偷税漏税的问题很重要，但是按他的估计其magnitude不足以改变其主要结论。