My goal is time series clustering, I would like to compare 5 minute returns, liquidity seems to be very low sometimes, so is it better idea to calculate average volume weighted price in 5 minute intervals and then divide one by another or rather should I use close prices calculated for 5 minute periods and divide them (standard approach) ? The problem is that, the last transaction could occur at 2020-09-07 09:07:00 for example, so this price is delayed by 3 minutes and real closing price of this 5 minute interval is unobserved. What would be the best practice here, is there relevant literature ?
$\begingroup$
$\endgroup$
-
1$\begingroup$ If you have defined your 5-minute periods as $[t,t+5)$, then I don't see how the closing price would be unobserved after the interval. You are unlikely to find a lot of literature about this -- though some of the high-frequency realized variance literature gets into it a little. What sort of clustering do you want to do? A 5-minute VWAP might be sensible or not. $\endgroup$ – kurtosis Sep 7 '20 at 20:21
