Data frequency and Asynchronism

This is from the CFAI text SS6 page 18. “Data of high frequency are more sensitive to asynchronism across variables. As a result, high frequency data tend to produce lower correlation estimates” I looked up the definition of asynchronism online but still don’t understand it. a lack of synchronism or coincidence in time

Think about 2 time-series data (i.e - stock returns) for each of the following: daily movements and monthly movements. The daily movements of 2 stocks will be noisy and volatile (i.e - 1 day it gives +0.4% return, the next -0.6%, etc). When you measure the correlation of the 2 stock returns using daily data, it will likely be very low. Monthly movements: this will be less noisy for the 2 stocks and more correlated to each other. In general, the longer out your data, the more “smoothed” it is. That is why high-frequency data is very noisy (which they call asynchronous). Also, because of this noise, correlation will be very low. Hope that makes sense.


Thanks, that makes sense now

MP - Is the below another example of this effect, or is it a Schweser over simplification of what you’ve explained above? Schweser (I’m dirty and used, I know) says that in the case where you only have access to a short time span of data, this might force your hand towrds high frequency. They say that in this event, because of mis-matched public holidays etc some days there is no data value for a given series and you / your data provider might plug this gap with a previous close figure, for example. If you then use this data for a regression, it will be be asynchronous. Thanks

I think it reveals the underlying issue between using more granular (high-frequency) data. It generalizes the difference between having a large data sample compared to a small one. Over a longer period of time, that close figure wouldn’t really impact the rest of the time series as it’s only 1 data point out of many, but if you have a small sample of data points, that 1 close figure can impact the data more severely. Generally, high-frequency data is harder to obtain, so you have a smaller sample. That’s how I think about it at least.