Datastream
beta calculations
Methodology
The derivation of Datastream betas is based on the
method described in 'Predictability of British Stock Market prices' by S.
Cunningham, Journal of Royal Statistical Society Series C, 1973.
Note:
For the purpose of this explanation, observations
under exceptional conditions (such as stocks traded for less than 2½, large
price changes and so on) are ignored. The sample used is in the normal case 60
months of monthly returns.
The implementation of this method uses logarithmic monthly returns which Cunningham found to fit slightly better than percentage returns. The basic model estimated is for each stock in a market or market list the following standard bivariate regression:
is the market
logarithmic return at time t, the logarithmic return for an equity, and the alpha
and beta are parameters to be estimated. Cunningham showed that the coefficients are effectively constant across all
stocks; they are important to this description only in that they are necessary
for the derivation of thefor a stock. The adjustment of the estimated
beta coefficients can be broken up into three stages:
Step 1
The following estimates are created for
each equity:
Ordinary least squares (OLS) estimate of beta
OLS estimate of alpha
Variance of regression residuals
The variance of the beta estimate is the standard expression
where T is the number of valid observations for the two variables and
Step 2
The average alpha and beta coefficients are calculated over J constituents
Average alpha
Average beta
Estimated variance of the theoretical beta value
Step 3
Finally, the beta estimate and correlation are
calculated for each stock. This is the value displayed on equity programs:
The adjusted beta is a weighted average of the ordinary beta and the average beta if the alpha estimate is close to the average alpha coefficient. The weight on average beta is then increasing in the residual variance, decreasing in variance of the theoretical beta value. The weight on the ordinary beta is increasing in market volatility.
The formula used for beta correlation is:
This is equal to the
ordinary correlation coefficient, given that the data sample is chosen as for
the estimation of the ordinary beta coefficient.
See also