volatility computation

[saminny]

Hello,

I am computing daily volatility as the standard deviation of the returns/movements at the top of the order book for a symbol.

that is,

midpoint = bid+ask/2
return = new_midpoint+old_midpoint / old_midpoint
volatility = std_dev (return..)

What do you think about this? Is there some source where I can backtest my computation?

For aggregate volatility numbers, I simply maintain a 20-day moving average the daily volatility numbers.

Any suggestions are most welcome.

thanks,
Sam

 

[chromosome]

I'm confused about your midpoint. Is this tick-by tick? Bid and ask at the close on a given day? Also, it is more common to use the natural log of the relative return. If you still want to use your midpoint idea:

return relative = ( midpoint / midpoint [last period] )

ln of return relative = ln( return relative )

volatility = standard deviation ( ln of return relative )

 

[saminny]

Thanks for your reply.

Yes, I am computing tick by tick returns and then their standard deviation to get the daily volatility.

2 follow-up questions:

why is "logarithm" of the return required in this case?

Do you think simply averaging daily volatilities to get aggregate volatility over a certain period (say 20 days) recommended? I am thinking we need to do something special to get the monthly o or 20-day average.

thanks,
Sam

 

[panzerman]

Quote from chromosome:

I'm confused about your midpoint. Is this tick-by tick? Bid and ask at the close on a given day? Also, it is more common to use the natural log of the relative return. If you still want to use your midpoint idea:

return relative = ( midpoint / midpoint [last period] )

ln of return relative = ln( return relative )

volatility = standard deviation ( ln of return relative )

Volatility = stddev(ln of return relative)*sqrt(365) or sqrt(252) if you prefer trading days over calendar days. Also, your standard dev. is calculated over a period of time, such as 20 days for monthly volatility, or 5 days for weekly volatility.