summarizes forecasts of earnings from more than 50
investmentfirms.Theymeasurethesquaredforecast errors
bymonthoftheyearandcomputetheratioofanalystforecast
errortotheforecasterrorfromtimeseriesmodelsofearnings.
They find that thetime series models actually outperform
analystforecastsfromApriluntilAugust,but underperform
themfromSeptemberthroughJanuary.Theyhypothesizethat
this is because there is more firm-specific information
available toanalystsduringthelatterpartoftheyear.The
other study, by O’Brien (1988),
5 compares consensusanalyst forecasts from I/B/E/S with
timeseriesforecastsfromonequarteraheadtofourquarters
ahead.Theanalystforecastsoutperformthetimeseriesmodel
forone-quarter-aheadandtwo-quarters-aheadforecasts,doas
well as the time series model for three-quarters-ahead
forecasts, and do worse than the time series model for
four-quarters-aheadforecasts.Thus,theadvantagegainedby
analystsfrom firm-specificinformationseemstodeteriorate
as the time horizon for forecasting is extended.
Invaluation,thefocusismoreonlong-termgrowthratesin
earnings than on next quarter’s earnings. There is little
evidencetosuggestthatanalystsprovidesuperiorforecastsof
earningswhentheforecastsareoverthreeorfiveyears.An
early study by Cragg and Malkiel
6 compared long-term forecasts by five investment
managementfirmsin 1962 and 1963 withactualgrowthover
thefollowingthreeyears,concludingthatanalystswerepoor
long-term forecasters. This view is contested by Vander
Weide and Carleton (1988),
7 whofindthattheconsensuspredictionoffive-yeargrowth
in the I/B/E/S is superior to historically oriented growth
measuresin predictingfuturegrowth. Thereis an intuitive