Interpreting Distributions of Cross-Sectional Return Statistics

Summary

The document poses a statistical question arising from a value-at-risk application to US industry portfolios. For each date, the researcher calculates statistics such as the mean, median, standard deviation, and kurtosis across industry returns, then observes each statistic over time. The key issue is whether the time series of these daily cross-sectional estimates should be called a sampling distribution and what an aggregate such as the mean of daily means represents.

No answer or method is provided, and the document reports no data or conclusions. The distinction matters: each daily statistic summarizes a cross-section, while its sequence over dates also reflects time variation and dependence in returns. Treating that sequence as a conventional sampling distribution would require assumptions about the data-generating process and the sampling design. The note is useful as a framing of a statistical interpretation problem, but it does not resolve the terminology or establish what inference is justified.

Key ideas

• Each date produces a cross-sectional estimate from the industry portfolio returns.
• Tracking those estimates over time creates a time series of cross-sectional statistics.
• The meaning of an average of daily cross-sectional means depends on the sampling and time-series assumptions.
• The document raises the question but provides no answer or empirical analysis.

Tags

# Sample distribution of cross-sectional statistics of returns


# Sample distribution of cross-sectional statistics of returns












Currently doing an application of VaR on sample of industry portfolios in the US. I have a matrix of $n$ industry portfolios with $m$ time-series observations. I calculate cross-sectionally (for each trading day) the mean,median, std.dev and kurtosis of the sample $(r_{m,1},r_{m,2},...,r_{m,n})$. For each cross-section sample, we have one estimator. For the whole time-period we have $m$ observation of the cross-sectional statistics.

My question is:

Is the distribution of these cross-sectional statistics (let $\hat{\theta_t}$ denote the estimator for the sample at time t) a sampling distribution?

If this holds, what information could be extracted from the sample distribution (e.g "the mean of the means")

Shown in full with attribution under the source's licence. Licence: CC BY-SA 4.0 (Stack Exchange)

This summary was written by Stratmill's research agent from the original; it is not a copy of the source.