Benjamin Graham 1946 Article - Being Right In Security Analysis

Benjamin Graham, the man widely recognized as the father of security analysis, wrote the following article in 1946. Though the market has changed dramatically since then, his approach to judging the success of an analyst’s recommendations remain just as valid today.

On being right in security analysis

The most interesting and important work of the senior ana­lyst leads up to and includes the recommendation that one or more common stocks be purchased. How can we tell whether such a recommendation has been right or wrong?

This seems like a simple question, but a really satisfactory answer is not so easy to find. When a department store buyer recommends the purchase of certain merchandise, he implies that all –Â or nearly all –Â of it can be sold at the standard markup during the current season. In most cases the soundness of such recommendations can be readily checked by the sequel. When a stock market analyst recommends the purchase of stock at 80, on the grounds that its technical action indicates an upward move is imminent, it should not be too difficult to check the “rightness” of such a proposal. Most of us would agree that for the market analyst to be proved right the stock must advance, say, not less than four points in not more than, say, 60 days.

But if a security analyst should recommend the purchase of United States Steel at 80, as “a good buy,” what criteria of corresponding definiteness can we apply to test his wisdom? Ob­viously we cannot ask that the stock go up four points in 60 days. Shall we require that it advance 10% in a year? Or that it do 10% better than “the general market”? Of that, regard­less of market action, the stock should meet certain require­ments with respect to dividends and earnings, over, say, a five-year period?

We have no scoring system for security analysts, and hence no batting averages. Perhaps that is just as well. Yet it would be anomalous indeed if we were to devote our lives to making concrete recommendations to clients without being able to prove, either to them or to ourselves, whether we were right in any given case. The worth of a good analyst undoubtedly shows it­self decisively over the years in the sum total results of his rec­ommendations, even though precise criteria for evaluating them be lacking. But it is unlikely that security analysis could de­velop professional stature in the absence of reasonably definite and plausible tests of the soundness of individual or group rec­ommendations. Let us try, tentatively, to formulate such tests.

We return to our assumption that a security analyst is now recommending that United States Steel common be bought at 80. If this is a sound recommendation, not only must it work out well in the market, but it must be based on sound reasoning also. For without such reasoning we may have a good market tip, but we cannot have a good security analysis. The reasoning, however, may take various forms, and the meaning of the recom­mendation itself will vary with the reasoning behind it. Let us illustrate by four alternatives:

1. Steel should be bought because its future earning power is likely to average about $13 per share (see C. J. Collins’ article; the Analysts Journal, July 1945, p 23).
2. Steel should be bought because it is fundamentally cheaper at 80 than is the Dow-Jones Industrial Average at 190.
3. Steel should be bought at 80 because next year’s earnings will show a substantial increase.
4. Steel should be bought at 80 because that price is far below the top figure reached in the last two bull markets.

Reason 1 implies that Steel will prove a satisfactory long pull investment. That does not mean necessarily that it will average earnings of $13 over the next twenty-five years, but cer­tainly over the next five years. If this analysis proves correct, the purchaser will have both satisfactory earnings and dividends and an undoubted opportunity to sell out at a good advance. The correctness of the analysis and the consequent recommendation can be proved only over a five-year period or longer.

Suppose that the same suggestion, with similar reasoning, had been made in January 1937, when Steel was also selling at 80? Would that analysis have been right? No, even though the stock promptly advanced 57% to 126. For in no five-year period since 1936 have the earnings averaged $7 per share, and the 1937-44 average was about $5 per share. The rise to 126 within 60 days did not establish the rightness of this analysis, any more than the decline to 38 in the following 12 months would necessarily have proved it to be wrong.

The recommendation to buy United States Steel because it is cheaper than the Dow-Jones Industrial Average (reason 2) would represent a valid and standard form of analytical argu­ment. It may or may not be coupled with the statement that Steel is attractive in its own right. In the former case, it would be equal to recommendation 1, plus the assertion that Steel is cheaper than other standard issues. But the analyst properly may recommend Steel common on a comparative basis only, without claiming that it is intrinsically cheap. In that case he will be proved right if Steel performs better than the average, even though it may not do well by itself. For example, if Steel declines to 70 within a year from now, while the Dow-Jones In­dustrials decline to 140, the comparative recommendation should be called right – provided (a) it was originally couched in comparative terms, and (fo) it was backed by plausible ana­lytical reasoning. Proviso (fo) would seem necessary in every case where a single recommendation is tested, in order to make sure that the Tightness is not due obviously to mere luck.

Recommendations to buy a stock for the main reason that next year’s earnings are going to be higher (reason 3) are among the most common in Wall Street. They have the advantage of being subject to rather simple tests. Such a recommendation will be right if both (a) the earnings increase and (fo) the price advances –Â say, at least 10% –Â within the next 12 months.

The objection to this type of recommendation is a practical one. It is naive to believe that in the typical case the market is unaware of the prospects for improved earnings next year. If this is so, the favorable factor is likely to be discounted, and the batting average of recommendations based on this simple ap­proach can scarcely be very impressive.

Steel should be bought at 80 because it sold considerably higher in the last two bull markets (reason 4). Is this a valid type of reasoning for security analysis? Opinions may differ on this point, but in any case we can readily tell if such a recom­mendation proves right. The stock must advance substantially —say, 20% at least—in the current bull market.

Conclusions

The preceding discussion leads to some general conclusions, which are put forward on a tentative basis and as a starting point for controversy:

(a) In most cases the rightness of an analyst’s recommenda­tion can be tested by the sequel, provided he indicates the type and basis of his recommendation.

(b) Different types of recommendation –Â even though they all may call for the same action; for example, to buy Steel at 80 –Â will be tested for Tightness in different ways.

(c) Where a recommendation is made on a group basis, only the group result should be tested. Individual issues may be expected to go counter to the group trend.

(d) Professional standards for security analysis require that all recommendations indicate clearly both the type of recommendation made and the kind of analytical reasoning on which it is based.

Analysts recommend bonds and preferred issues as well as common stocks. The warranty behind each such suggestion is that the issue has quality at least commensurate to the yield. Such recommendations may be tested two ways: either by the review of the analysis of quality, or by the subsequent market action of the issues approved, preferably as compared with a suitable group index. This field of activity does not raise seri­ous problems of testing, except where very refined results are required.

Assuming we can test the analyst’s performance on indi­vidual recommendations, we can develop a crude batting aver­age for his work, based solely on the percentage of times he is right out of total number of recommendations made. How high should this average be for a good analyst? And is it necessary to refine this test, by distinguishing between “very right” or “very wrong” and just “rightish” or “wrongish”?

These are questions for others to answer.

H/T - http://post.nyssa.org/nyssa-news/2011/06/from-the-arcives-on-being-right-in-security-analysis.html