Comparing Market Betas with Incomplete and Unequal Return Histories

Summary

The document raises a question about estimating betas for three real estate markets relative to a national market when each market has returns for different periods. It asks whether all beta estimates should use only the periods shared by every series, so the resulting estimates can be compared on a common sample, or whether each pair should use its own overlapping observations.

No answer or method is provided, so the document does not establish which approach is preferable or give evidence about the resulting estimates. It identifies a real comparability issue: beta estimates can reflect both market relationships and the time period sampled. Restricting every pair to common dates improves consistency of the comparison but discards observations, and the example explicitly sets sample-size concerns aside. Any conclusion would require considering the research objective and the effects of the shorter sample.

Key ideas

• The document asks how to compare market betas when return histories cover different periods.
• Using a common set of dates makes pairwise beta estimates comparable over the same sample window.
• Estimating each pair on its own overlapping dates retains more observations but mixes different time periods.
• The text poses the issue but provides no answer or empirical evidence.

Tags

# Real estate returns with incomplete dataset?


# Real estate returns with incomplete dataset?












I have a dataset of real estate returns in markets A, B, and C. I also have national returns, denoted market N. I have 10 columns representing the time period for each data point. Market A only has data for periods 1 through 5. Market B only has data for periods 3 through 7. Market C only has data for periods 1 through 9. Market N has data for all periods. Therefore, the only periods where we have data points for ALL markets are periods 3, 4, and 5.

With the above in mind, I want to calculate the betas for each market pair: AN, BN, CN. Disregarding the size of the sample (illustrated as such for simplicity), would it make most sense to calculate beta for each pair using data only from periods 3, 4, and 5? I want to compare betas to each other. I initially calculated beta for each pair based on which periods they themselves as a pair had in common but then realized this may not make sense. If each beta is calculated with different periods, then they are not comparable. Even though we lose out on data the way I explained above, at least we lose out on the same periods across each market pair. Am I making sense? Thank you!

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.