Weighting Monthly Returns for CAPM Correlation Estimates
Summary
The question asks whether monthly portfolio and benchmark returns should be adjusted for differences in month length when estimating annual excess returns, correlations, and CAPM betas. One response suggests normalizing by trading days where possible. Another says such adjustments are not commonly seen and argues that irregular month lengths may matter less than the choice between monthly, weekly, or daily sampling.
The discussion does not provide a tested estimator, empirical comparison, or definitive consensus. It notes that irregularly sampled time series are treated in other fields, but the cited research is not assessed by the respondent. The practical takeaway is that day-count weighting is a possible consideration, while the available answers offer limited support for routine adjustment and do not resolve which weighting convention is best for a particular dataset.
Key ideas
- Monthly returns have unequal calendar and trading day intervals.
- One answer recommends normalizing observations by trading days when feasible.
- Another answer reports that month length adjustment is not common practice.
- The discussion suggests sampling frequency may matter more, but gives no empirical test.
Tags
Full text
# different amount of information on return correlations from shorter and longer periods? # different amount of information on return correlations from shorter and longer periods? I want to calculate annual excess returns on portfolios using monthly (total) returns for a CAPM (for the assets in the portfolio as well as for the benchmark), in order to have more information on the correlations, more precise betas. Is it standard practice to adjust (slightly) for shorter months having somewhat less information on the correlations? Shall I weight by the number of days or only trading days before return dates? Full disclosure: This breaks down my longer question into specifics. Please bear with me. From: annual excess returns from CAPM on monthly total returns ## Answer by SCallan (score 1) https://quant.stackexchange.com/a/10492 Normalize for trading days if possible. ## Answer by deprecated (score 1) https://quant.stackexchange.com/a/11563 I have never seen such an adjustment. While monthly data are irregularly sampled in time (in every way...calendar days, trading days, seconds, etc), that irregularity is likely to be a smaller effect than your choice of data frequency (monthly, weekly, daily data). That said, your question is intriguing because in other fields they do have to deal with irregularly sampled time series data and methods (apparently) exist. See this paper for example. This is way outside of my field so I can't vouch for the paper but the issue is interesting.
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.