Missing the Barn Door

Im sure that many readers know only too well what happens when a stock misses the consensus forecast by much. High flying 3Com tumbled 45% when analysts forecasts missed reported eamings by a scant 1 % in 1997. Sun Microsystems dropped 30% on a 6% shortfall. On May 30,1997, Intel announced eamings for its June quarter would be sharply higher than the 1996 quarter; however, they would be somewhat below the analysts consensus forecast. (It tumed out by 3%.) This caused a price drop of 26 points or 16% on the opening, which reverberated first through technology stocks and then through the market as a whole. The result was the S&P 500 lost $87 billion in minutes. Oak Technology, a high-flying IPO, dropped from $32 in March to $10% in early June 1996, primarily as the result of disappointing earnings. Cascade Communications, then a leading computer networking company, plummeted from $91 in late 1996 to $24 by March of 1997, or 76%, as a result of earnings falling short of the analysts consensus estimate. Merrill Lynch and Lehman, at the time, cut their estimates by about 8%. Ditto Zoll Medical, which fell 26% on a 1-cent eamings shortfall to expectations. People who relied on these estimates got clobbered-to say the least.

But these are horror stories. Again, we must ask if they are exceptions. How good is this dedicated and hardworking group at their vocation? To answer this question, we looked at how close forecasts came to

four different ways.° In all cases, errors were high. What about surprises from companies that report small or nominal eamings? Wouldnt a miss here result in a higher percentage error than one from companies reporting large eamings? If, for example, the estimates were $1.00, and the company actually reported 90 cents, the miss would be 10%. But if the estimates were 10 cents and the company reported only 1 cent, the miss would be 900%.

To measure this effect, we eliminated all companies that reported eamings in the -i- or - 10-cent range to prevent large percentage errors from this group distorting the study. The problem is that many of the fastest growing small companies report eamings of 30 to 50 cents a share annually, which translates into 7.5 to 12.5 cents quarterly. Large companies in this range were also eUminated. Even using this ultra-conservative method, the average forecast error was still 23% on average, more than quadmple the size that market pros believe could set off a major price reaction.

Figure 5-2

The Forecasting Follies

Percentage of estimates inside of range

±5%

± 10%

± 15%

Percentage (plus or minus) by which analysts hit actual eamings

1973- 1996 Sample Size: 94,251 Estimates

Source of data: A-N Research Corp. (Formerly the research department of Abel Noser .) and I/B/E/S

levels considered acceptable to the pros. We have seen that eamings surprises of even a few percentage points can trigger major price reactions. Current investment practice demands estimates that are very close to- or dead on-reported results. Normally, the higher the valuation of a stock, the more important the precision. Street-smart pros normally expect reported eamings to be within a 5% range of the consensus estimate-and many demand better.

Is it doable? Look at Figure 5-2. This time we examined the number of consensus estimates that fell within the ranges most professional investors beheve will have no impact on the stock price. We used our large database of nearly 100,000 analyst consensus estimates for the 24 years to the end of 1996.

The figure summarizes what we found. Again, the resuks are devastating to believers in precise forecasts. The distribution of estimates clearly refutes their value to investors. Only about one in four was in the ±5% range of reported eamings that a good number of pros deem es-

sential. Using the plus or minus 10% error band, which many professional investors would argue is too large, we found that only 47% of the consensus forecasts could be called accurate. Almost 55% missed this more lenient minimum range. Worse yet, only 58% of the consensus forecasts were within a plus or minus 15% band-a level that almost all Wall Streeters would call too high.

Of what value are estimates that seriously miss the mark two-thirds or three-quarters of the time? After the horror stories precipitated when forecasts were off even minutely, the answer seems to be ... not much. We have seen then that estimates carefully prepared only three months in advance, by well-paid and diligent analysts, are notoriously inaccurate. To complicate matters, many stocks sell not on todays earnings, but on expected eamings years into the future. The analysts chances of being on the money with their forecasts are not much higher than winning a lottery. Current investment practice seems to demand a precision that is impossible to deliver. Putting your money on these estimates means you are making a bet with the odds stacked heavily against you.

"But, maybe, there is a reason for this," believers in their forecasting prowess might argue. "Analysts may not be able to hit the broad side of a bam overall, but thats because there are a lot of volatile industries out there that are impossible to forecast accurately. You can make good estimates where it counts, in stable, growing industries, where appreciation is almost inevitable."

A plausible statement. We fed it to our computer, which digested the database and spit out the answer a few minutes later. We divided the same analysts consensus estimates into 62 industry groups and then measured accuracy for each. The answer is shown in Table 5-1. Alas, the results were no better. The average error was 50% and the median 43% annually. These errors were a touch larger than forecasting individual companies without considering the industries they are in. We also found that over the entire time period, 90% of all industries had analyst consensus forecast errors larger than 20% annually, while 10% of industries showed surprises larger than 86%. As the chart indicates, analysts errors occurred indiscriminately across industries. Errors are as high for industries that are supposed to have clearly definable prospects, or "visibility," years into the future, such as computers or pharmaceuticals, as they are for industries where the outlooks are considered murky, such as autos or aluminum.

We also examined the standardized error, a statistical tool that adjusts for the volatility of reported eamings for differing industries. We found that after adjusting for the volatility of various industries, forecasting er-