Overlapping Returns and Bootstrap Inference in Crash Analysis

Summary

The document raises methodological questions about an analysis that groups starting observations by monetary-distortion index quartiles, then examines the lower tail of subsequent two-month returns. The cited procedure uses overlapping two-month returns within a three-year window and reports the second and fifth percentiles for each bucket. The author asks why returns overlap and whether the three-year windows overlap as well.

Overlapping observations can provide more possible return periods within a limited sample, but adjacent returns share data and therefore are dependent; standard errors or bootstrap procedures must account for that dependence. Non-overlapping returns reduce this overlap but leave fewer observations and can make results sensitive to the chosen starting dates. The document itself does not answer these questions, describe the bootstrap implementation, or present underlying data or findings beyond the quoted methodological setup. Its skeptical comment about the book’s framing also signals that the analysis may depend on strong prior assumptions, so the sampling choices and interpretation need independent scrutiny.

Key ideas

• The cited analysis buckets starting observations by index quartile and examines subsequent two-month return tails.
• Overlapping return periods increase the number of observations but create dependence between neighboring samples.
• Bootstrap standard errors should reflect dependence introduced by overlapping returns.
• The document asks whether the three-year sampling windows overlap but supplies no answer or implementation details.

Tags

# Application of bootstrap to bucketed returns data


# Application of bootstrap to bucketed returns data












In the (wonderful) book "The Dao of Capital", the author, M. Sptiznagel, describes at pag. 242 this procedure to estimate the crash losses and bootstrap standard errors that follow high MS Indices:

> Upon bucketing two-months returns by their starting MS index quartiles (over a 3 year window of overlapping two month returns following bucketing) and calculating the 2nd and 5th percentiles in each bucket, we see, again, that crashes follow distortion.

Since he uses this kind of framework of data analysis several times in the book, I have some questions related to the data construction:

- why taking overlapping, instead of non-overlapping, 2-month returns within the 3-years window?

- do you think the 3-years windows, in which I guess the author is looking for a crash that will "adjust" the monetary distortion, are overlapping too? If not, why?

I understand that the author is doing the data analysis part with the "clock ticking" (p.229) since the aprioristic conclusions of the Austrian School are correct despite empirical results! Anyway, I would like to discuss pros and cons related to the 2 questions.

Let me know if more details are needed. Thanks.

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.