relationshipamongandwiththeindependentvariablesused
in the regression.
- Thefactthatmultiplesarenot normallydistributed
canpose problemswhen using standardregression
techniques. These problems are worse with small
samples,wheretheasymmetryinthedistributioncan
be magnified by the existence of a few large outliers. - In a multiple regression, the independentvariables
arethemselvessupposedtobe independentof each
other.Consider,however,theindependentvariables
that we have used to explain valuation
multiples—cash flow potential or payout ratio,
expectedgrowth,andrisk.Acrossasectorandover
the market, it is quite clear that high-growth
companies will tend to be risky and have low
payouts. This correlation across independent
variables creates so-called multicollinearity, which
can undercut the explanatory power of the regression. - Earlier in the chapter, we noted how much the
distributionsformultipleschangedovertime,making
comparisonsofP/EratiosorEV/EBITDAmultiples
across time problematic. By the same token, a
multipleregressionwhereweexplaindifferencesina
multiple across companies at a point in time will
itselflosepredictivepowerasitages.Aregressionof
P/E ratios against growthrates in early 2005 may
thereforenotbeveryusefulinvaluingstocksinearly
2006. - Asafinalnoteofcaution,theR-squaredonrelative
valuation regressions will almost never be higher
than 70 percent, and it is common to see the
R-squareddropto 30 or 35 percent.Ratherthanask