If analysts could forecast earnings changes perfectly, they could forecast stock prices. Most empirical evidence, however, supports the conclusion that analysts cannot forecast earnings with enough accuracy to improve upon an extrapolation of past growth rates, which is itself not a useful predictor.

Professional investors can extrapolate past earnings trends fairly successfully. A large portion of estimates based on extrapolation turn out to be right, enforcing the impression that such forecasts have value. In fact, however, most of the value for anticipating stock price changes lies in forecasting changes in earnings trends. When investors’ earnings forecasts are close to the mark, they mean very little to the price of the stock; when they are wrong, they mean a great deal.

Most of us do not recognize that it is the exception, rather than the rule, that we know more than the market. Tending to treat all forecasts with the same degree of seriousness, we end up investing in those companies with superior earnings records while ignoring those without. Expecting the future to continue on a straight line from the past, with no surprises in store, we tend to pay too much for companies with high price-earnings ratios and too little for low P/E companies.

The analyst who paid $120 a share for Atlantic Richfield in 1969 would have laughed out of Wall Street anyone who told him that the stock would fall to $48 in a year’s time because the caribou’s migrations would require redesign of the Alaska pipeline. The “caribou” eventually visit most stocks, and when they visit one with a very high P/E multiple the consequences can be severe. Yet analysts continue to ignore the great potential of low P/E stocks in favor of a set of forecasts extrapolated from past achievements.