I am doing some academic work and using 1 minute bar data. I am wondering if when calculating the return time series, do I need to throw out the first return of the day because it is the return calculated from first bar today and last bar yesterday? My reasoning is that this return is not calculated over 1 minute time frame. It makes sense to me to throw this return out as "atypical".

for example you can see the jump from 2017-01-06 to 2017-01-08 here (6th being Friday and 8th being Sunday - this is an FX market)

df = data_fx['EURUSD']
df[df.index >  '2017-01-06 21:55:00'].head()

2017-01-06 21:56:00    1.053315
2017-01-06 21:57:00    1.053320
2017-01-06 21:58:00    1.053455
2017-01-06 21:59:00    1.053380
2017-01-08 22:00:00    1.053050
2017-01-08 22:01:00    1.053040
Name: close, dtype: float64
  • $\begingroup$ The general answer is to just use bars that really were containing the one minute difference between the two prices. But your data seems weird ? If it's another day, then why does it look sequential time wise ? Shouldn't the time of the eigth day be not sequential with the sixth day ? $\endgroup$ – mark leeds Aug 1 at 18:24
  • $\begingroup$ Thanks - the data is one minute bars between market close and open. I've chosen a poor example possibly - this represents the weekend (6th being Friday and 8th being Sunday when FX market opens). In any case this further illustrates my point - you get even a larger gap between last point on Friday and first point after the weekend. This is no longer a 1 minute bar. $\endgroup$ – s5s Aug 1 at 18:50
  • $\begingroup$ It depends what you are trying to do. The "overnight return" is different from the other 1 minute returns, as you pointed out. So you can exclude it and use only intraday returns, or you can separately handle overnight returns and intraday returns to get the complete picture. $\endgroup$ – noob2 Aug 1 at 21:29
  • $\begingroup$ @s5s: as noob2 said, it just depends on whether your model is intraday or intraday + overnight. In other words, are you making predictions in both periods or just intraday. if just intraday, then don't include anything connected to the overnight returns. $\endgroup$ – mark leeds Aug 2 at 4:01

You could ignore the problem, throw away returns for the first minute in each trading session, or you could keep those first-minute returns in a second dataset and analyze that to estimate the overnight gap volatility.

Also, since you are an academic and using minute bars, I'm sure you are aware of the issues with bid-ask bounce and volatility estimation for frequently-sampled prices. If not, look for work by Andersen on volatility signature plots as well as work by Aït-sahalia, Mykland, and Zhang plus Podolskij and Vetter.

| improve this answer | |

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.