Interesting question.

Academic research shows that merely taking the average earnings of the last 8 years of EPS works remarkably well.

In an ideal world I might do this:
Fit a regression line through the earnings per share.
Find out which EPS figure is the furthest from the line.
If it's not the most recent one, replace it with a point on the trend line.
The thinking is that even the best companies are allowed one outlier year!
Then, do the regression over again and measure the R-squared quality of fit.
The better the fit, the steadier the earnings.

This process is probably more useful for finding an estimate of
cyclically adjusted earnings: what this year would be like if it
were neither unusually good nor unusually bad.
If you want the speed of the trendline of earnings growth, replace all
the earnings with their logarithms before doing the trend line.
Then the slope of the trend line will be commensurable between
companies with different absolute levels of EPS.

I would tend to use figures including extraordinary items.
Adjusted earnings without those anomalies might be a truer guide in
any given year, but over time they add up and they are real.

One way I have used the Value Line figures to estimate forward earnings:
Start with the cyclically adjusted earnings figure for this year.
Increase it by an annual rate of increase that is the lowest of
- Book value growth per share last 5 years
- EPS growth last 5 years
- Projected EPS growth rate
- Some sane constant to set a cap. 10-12-15% maybe?

Then, estimate earnings 8 years from now as current EPS * (1+growthrate)^8

This sort of pure-quant analysis would be good for finding candidates,
but a lot more examination would be in order. Sometimes a firm with
preternaturally smooth earnings is actually a risky firm.

Jim