Higher or lower Correlation estimate?

Which one(s) of the following will produce a lower correlation estimate? Why?

1, Appraisal(smoothed) data 2, high-frequency data 3, stale-price bias


PS. More than one choice, I think.

I would say appraisal method.For assets that have no liquid/deep markets, it is difficult to estimate parameters such as return/volatility given infrequent market transactions, motivating an appraisal approach to valuation…this smoothing approach has a tendency to underestimate volatility aswell as the correlation of the asset’s returns with the returns of other asset classes.

I would think

a. Appraisal (smoothed) data -> data is being smoothed for lack of pricing data. So this would be a lower estimate than if the prices were available. Here volatility would be reduced due to the smoothing.

b. Stale Price Bias - older stale prices - would also have pretty much the same effect as above. Lack of prices = Lower volatility = lower correlation estimate.

All 3 will bias correlation downward. High frequency data is prone to asynchronysm across variables which causes artifical low variance, sd, and correlations.

Appraisal data and stale pricing is inter-related as the information between appraisals can be old (stale) causing correlations and volatility measures to appear lower than they actually are.

I would say stale price bias only.

Smoothed data and high frequesncy data I believe both INCREASE correlations artificially.

My answer was not a guess, i know it is correct with absolute certainty…

Can you cite the book/reading # for High frequency data? I remember reading this but cannot find it now.

Last years mock exam - afternoon. Number 26 deals with appraisal data and 27 with high frequency. I knew this because I did it (again) like the day before you posted this.

high-frequency data?

High frequency daily returns have lower correlation by a higher standard deviation.

It is strange but true that both higher frequency data NAD lower frequency date (appraised values, stale prices) produce lower correlations. I guess you need just the right amount of frequency.

so all three lower the correlation estimate. We can not determine however which produces the lowest amongst them.

markFAIL is correct. I will review this again in late may.

  1. stale price bias is on page 74. Measured standard deviation could be lower or higher. (Another mystery to me, can you explain it in plain language?) Thanks.

prices are not changing. If the actual prices in the market had they been available were lower - your standard deviation shown by the stale prices is higher. But if the prices had they been available were higher - your standard deviation calculated with the stale prices is lower.