- this is the a general description of a measure; if this does not comply with Statalist rules, feel free to delete it. My question is in the second post of this thread. The question can be understood without reading the first post, hopefully! -
I am going to go quickly highlight the main concept of the Roll bid-ask spread estimator, as this thread might be useful for other finance students. This estimator of the bid-ask spread shall reflect the liquidity of a security, given that some assumptions hold (see: Roll, The Journal of Finance, 1984). It can be used for low frequency data, basing on last prices, or on high frequency, basing on intraday data.
This liquidity measure can be estimated as follows:
spread estimate = 2*sqrt(- covariance (price_change_t+1, price_change_t)) // in the case the spread should be expressed as percent, replace 2* with 200*
The Roll measure can be calculated for arbitrary intervals. For daily estimates, a full trading month of data, i.e. 21 trading days, are recommended and the daily estimator should be calculated on days with at least one transaction. To be well defined, four transactions should take place within that 21-day period (more relevant for corporate bonds, than for stocks). The serial covariance is then calculated based on the price changes in that 21-day period. For daily estimates, a rolling window of 21 days is useful. For quarterly estimates, the median of the daily estimates may be used.
I am going to go quickly highlight the main concept of the Roll bid-ask spread estimator, as this thread might be useful for other finance students. This estimator of the bid-ask spread shall reflect the liquidity of a security, given that some assumptions hold (see: Roll, The Journal of Finance, 1984). It can be used for low frequency data, basing on last prices, or on high frequency, basing on intraday data.
This liquidity measure can be estimated as follows:
spread estimate = 2*sqrt(- covariance (price_change_t+1, price_change_t)) // in the case the spread should be expressed as percent, replace 2* with 200*
The Roll measure can be calculated for arbitrary intervals. For daily estimates, a full trading month of data, i.e. 21 trading days, are recommended and the daily estimator should be calculated on days with at least one transaction. To be well defined, four transactions should take place within that 21-day period (more relevant for corporate bonds, than for stocks). The serial covariance is then calculated based on the price changes in that 21-day period. For daily estimates, a rolling window of 21 days is useful. For quarterly estimates, the median of the daily estimates may be used.
Comment