## A Review Of Ben Graham’s Famous Value Investing Strategy: “Net-Nets”

Benjamin Graham, often considered a strong candidate for the “the father of quantitative value investing“, developed an investment strategy that involved purchasing securities for less than their *“current-asset value”*, *“a rough index of the liquidating value”. *We uncovered ten research papers that examined the returns achieved by investing in such securities which were conducted over a number of decades and across various geographies. In general, the research found that the strategy generated a remarkable level of outperformance. Based on what we know about the “publication bias,” perhaps this not too surprising.

But our objective in reviewing this collective research was not to identify if the Graham “Net-net” strategy works from a historical perspective, instead, we seek to answer the following question, “Could the returns reported in the research have been achieved by an investor in practice?”

To meet this objective, we developed a methodology to analyze the evidence and determine its reliability. Subsequently, we found that each of the studies suffered, or may have potentially suffered, from a number of biases that adversely impacted the reliability of the results contained therein. We concluded that a practitioner would be *unlikely to earn the returns reported* in the research. Lastly, we briefly discussed the implications of our findings for practitioners of the investment strategy, as well as for evidence-based investors at large.

### The History of the Net-Net Value Investing Strategy

In 1934, between two world wars and in the midst of the great depression came the publication of Security Analysis by Benjamin Graham and David Dodd, an investment classic.

“A classic is something that everybody wants to have read and nobody wants to read.”attributed to Mark Twain

Within the weighty tome Graham details an investment strategy that involves purchasing stocks for less than their *“current-asset value”*, *“a rough index of the liquidating value”*. ^{1} In turn, the *“current-asset value of a stock consists of the current assets alone, minus all liabilities and claims ahead of the issue. It excludes not only the intangible assets but the fixed and miscellaneous assets as well.”* ^{2} Colloquially, such firms are referred to as “net nets” because their market capitalization is “net” of the “net current asset value”. ^{3}

For context it is worth noting that the Dow Jones Industrial Average (DJIA) reached 381.17 on September 3, 1929 and bottomed at 41.22 on July 8, 1932 – an 89.2% drawdown! ^{4} One can only marvel at the intestinal fortitude demonstrated by implementing *any* stock investment strategy in the face of such capital destruction.

Furthermore, with the behaviour of the market such as it was in the lead up to the publication of Security Analysis, one wonders how Graham’s faith was maintained in the behaviour of market participants, given the ultimate reliance on market participants to achieve the desired capital appreciation.

“Successful investing is having everyone agree with you… later”Jim Grant

Graham is often considered to be the father of value investing. However, despite such reverence, his strategy of purchasing securities trading at less than their Net Current Asset Value (NCAV) has been the focus of relatively limited research. In our quest for research examining the returns achieved by purchasing such securities, we uncovered ten studies that were conducted by academics and practitioners alike.

### The Objective

Our analysis was focussed on answering one *primary* question:

**Could the returns reported in the research have been achieved by an investor in practice?**

At first glance it seems an almost preposterous research endeavour; why examine something that is seemingly self-evident? Well, the returns reported in the majority of the research revealed an outperformance of such magnitude that our inherent scepticism stirred us into forensic examination.

“[I]n theory there is no difference between theory and practice, while in practice there is”

Benjamin Brewster (often attributed to Yogi Berra)

### The Methodology

In order to meet our research objective and answer the question, “Could the returns reported in the research have been achieved by an investor in practice?”, we systematically and objectively analysed each study in its own right. Accordingly, we analysed a cross section of items in each of the studies including, but not limited to, the: valuation metric, weighting method, purchase/rebalancing rules, portfolio formation methodology and holding period(s). With regard to reliability specifically, we analysed the studies for: survivorship bias, look ahead bias, sample size issues, potential for human error, publication credibility, time period bias, data source reliability, minimum market capitalization requirements, return calculation methodology and errors generally.

To enhance understanding we provide additional detail pertaining to certain items below:

### Time period bias

In general, the longer the test period the more reliable the results, the shorter the test period the less reliable the results. That said, where individual studies are deemed reliable, but for the time period utilized, we could potentially marry them to other such studies and aggregate the results in an attempt to circumvent the time period bias.

According to commentary and research published in “What Works on Wall Street (Fourth Edition)” by James O’Shaughnessy investing the lowest (i.e. cheapest) decile of stocks sorted by price to book has produced material outperformance for periods as long as 18 years. However, over even longer periods that relationship has not held. Ideally then, we would want multi-decade examination periods to be employed. (here is a simulation study on value portfolios to add additional context)

We used the following ranges to determine the presence of time period bias:

- < 10 years; inadequate/unreliable
- 11 to 20 years; somewhat reliable
- 20 years; more reliable
- 40 years; most reliable

### Data source reliability

A reliable and reputable data source is a necessity when conducting empirical research. Seemingly, the utilization of the CRSP/Compustat data base covering US listed securities from 1926, the gold standard in stock research, would circumvent any concerns when examining US listed securities. Unfortunately, even this database has had concerns raised over its reliability in a highly recommended article titled “The Myth of 1926: How Much Do We Know About Long-Term Returns on U.S. Stocks?” by Edward F McQuarrie. ^{5} McQuarrie finds that, in essence, 1973 onwards represents the point at which data contained within the CRSP/Compustat data base is most reliable. However, in a final twist James O’Shaughnessy states in “What Works on Wall Street (Fourth Edition)” that *“[c]ompustat also added many small stocks to its dataset in the late 1970s that could have caused an upward bias to result, since many of the stocks added were added because they had been successful.”*

Consequently, if even the leading database for empirical research into stocks has question marks over its reliability, we would be wise to assume that other databases may also suffer from some form of bias or contain errors of some degree.

### Minimum market capitalization requirement

For an investment strategy to be effective, it must be tradable in practice. However, some studies do not mandate a minimum market capitalization requirement for the stocks contained in the investment universe under examination. Consequently, the results in such studies can be unduly influenced by stocks that are, in practice, virtually untradeable, even when attempting to deploy relatively modest amounts of capital.

To be clear, we do not refer to mere micro capitalization stocks, despite their identification as a major source of the proliferation of “anomalies” identified by academics. ^{6} In fact, we are not even concerned with stocks close to the upper bound of “nano capitalization” classification per se. ^{7} Rather, our concern lies with the very smallest stocks that trade infrequently and at a very low “dollar” volume.

We have seen empirical evidence where the smallest decile of stocks was reported to have generated a Compound Annual Growth Rate (CAGR) of 84%! ^{8} Of course, the driver of those returns were almost certainly stocks that were uninvestable in reality. We explore this issue further by referring once again to “What Works on Wall Street (Fourth Edition)” by James O’Shaughnessy. O’Shaughnessy found that between 1964 and 2009 stocks with a market capitalization less than a deflated USD 25 million (2009 dollars) generated a CAGR of 63.2%! ^{9} However, *“when you require that all stocks have share prices of greater than $1, have no missing return data, and have limited the monthly return on any security to 2,000 percent per month”* the CAGR fell to 18.2%. Furthermore, when the analysis was extended to 1926 he found that the CAGR fell to 15%. And finally, *“[w]hen you look at the results for investable microcap names, those with market capitalizations between a deflated $50 million and a deflated $250 million, you see that most of the return for tiny stocks disappears.”*

Incorporating the principle identified above, where research contains (or may contain) stocks that are (or likely to be) untradeable we may refer to it as suffering from **“uninvestable stock bias”**.

### Return calculation methodology

To achieve our objective of answering the question, “Could the returns reported in the research have been achieved by an investor in practice?”, the answer was reliant on the return methodology adopted to quantify the reported returns.

What follows is a list of the various terminology used to describe the returns in the studies examined: annual geometric mean return, mean returns, abnormal performance, abnormal return, average raw buy-and-hold, average market-adjusted buy-and-hold, abnormal buy-and-hold performance, average return p.a., market return, raw returns, market index returns, buy-and-hold raw return, market-adjusted return, average raw return, average return, compound annual growth rate, cumulative raw return, excess return, percentage of positive excess return, cumulative excess return, simple average, and annualized return!

While the jargon is mind boggling, in general, the returns reported in the research generally refer to two simple return calculation methodologies: the arithmetic mean and/or the geometric mean.

**In a dependant return series that exhibits volatility (like stock returns) the arithmetic mean will, as a matter of mathematical law, overstate returns relative to the geometric mean.** ^{10}

For ease of reference, where returns have been calculated using the arithmetic mean we may simply refer to the research as suffering from **“inflated return bias.”**

For clarity, it should be noted that the geometric mean, when calculated for annual periods is also often referred to as the Compound Annual Growth Rate (CAGR) or the “annualized” return. The geometric mean return represents the *actual return potentially achievable by an investor in practice*, and therefore it is the sought-after measure when quantifying investment returns.

Given the importance of the return calculation methodology a few rudimentary examples are worthwhile. Firstly, let us assume an investor starts with $100 and they incur a 60% loss after 1 year resulting in a portfolio value of $40 ($100 * (1 – 60%)). Then, in year 2 they generate a 100% gain resulting in a final portfolio value of $80 (40 * (1+100%)). The arithmetic average return in this case would be 20% ((-60% + 100%) / 2)! Clearly, the result is nonsensical to an investor in practice. In contrast, the geometric mean would reflect practitioner reality and yield a result of -10.56% (80/100^(1/2)-1).

Another theoretically extreme example is provided in the table below:

The results are hypothetical results and are NOT an indicator of future results and do NOT represent returns that any investor actually attained. Indexes are unmanaged, do not reflect management or trading fees, and one cannot invest directly in an index.

Remarkably, it is theoretically possible to achieve an arithmetic mean twice that of the market (16.00% vs 8.00%; 2x); and simultaneously attain only a fraction of that return based on the geometric mean (4.99% vs 0.95%; 0.19x).

Finally, we provide a practical and wholly independent example to illustrate the misleading nature of an arithmetic mean when dealing with investment returns. Accordingly, we reproduce below Table 1 from the “Summary Edition Credit Suisse Global Investment Returns Yearbook 2019”:

The results are hypothetical results and are NOT an indicator of future results and do NOT represent returns that any investor actually attained. Indexes are unmanaged, do not reflect management or trading fees, and one cannot invest directly in an index.

There is a material difference between the arithmetic and geometric mean achieved for all equity markets examined. For Japan, the arithmetic mean is more than double the geometric mean!

Notwithstanding that discussed above, one may still be tempted to use the arithmetic mean as a guidepost to estimate the more practically meaningful geometric mean; we would caution against such an endeavour. In addition to the mathematical pitfalls illustrated in our examples above, psychologically speaking, such thinking may be driven, in part, by confirmation bias (“net nets outperform!”) and sunk cost fallacy (i.e. having put in the time and effort to read a study one may *want* to walk away “knowing something definitive”). In addition, statistically, as the number of holdings in a portfolio falls (an issue when examining the relatively small universe of firms trading below NCAV) the volatility of that portfolio may increase thereby leading to a greater potential divergence between the geometric and arithmetic mean. How portfolio volatility changes with the number of holdings in a portfolio was examined, for example, by Elton and Gruber in “Risk Reduction and Portfolio Size: An Analytical Solution” and by Alpha Architect here and here. ^{11}

So, mathematically, psychologically and statistically attempting to estimate the geometric mean is precarious and even more speculative than it may initially appear.

“The first principle is that you must not fool yourself and you are the easiest person to fool.”Richard P. Feynman

This leads us to an obvious question, “*Why* would you use an arithmetic mean to calculate stock returns?”. Indeed, when it first dawned on us that such a methodology was used to quantify stock returns, we were left utterly dumbfounded. We asked finance academics why they utilize the arithmetic mean, and the primary reasons given were that it is:

- Required for the statistical methods applied in academic research e.g. regression analysis
- Used to calculate risk measures e.g. Sharpe ratio, standard deviation etc.
- Used to circumvent the effect of the start and end date which may unduly influence the returns

While the above serves a purpose in *academic research* it does little for an investment practitioner trying to determine how much they can earn on their capital in reality. ^{12} While on occasion academic research does contain the geometric mean return (i.e. CAGR) it is not, in our experience, a common occurrence. ^{13} Indeed, most academic finance research appears to be conducted using arithmetic mean returns. Simply including the CAGR alongside the arithmetic average within academic research would greatly enhance its utility to the practitioner community, and we hope the measure is increasingly adopted. That said, it is our understanding that the academic community rejected the notion of “maximum drawdown” as a measure of “risk”, despite it being arguably the most important risk-related metric for a practitioner. ^{14} Consequently, the adoption of seemingly “common sense” measures may not be as inevitable as one would hope. Indeed, there appears to be a significant gap between the ivory tower and practitioner land, consequently, we are probably all the poorer as a result.

While our overall methodology used to examine the various studies may seem onerous, we believe it was required as all too often we have seen the cognitively deficient assertion that “all the evidence about “x” says “y” without any thorough examination of underlying evidence, in and of itself. It goes without saying, no matter the quantum of studies showing the same or similar results they remain collectively worthless if they each contain material methodological flaws – they ought to be dismissed rather than “anchored” to in the mistaken belief that they possess some utility. ^{15}

It is often said “I’ve never seen a bad back test”. We think it prudent to amend that quote to, “I’ve never seen *bad back test results!*”.

# The Evidence

The table below lists the research papers examined along with their key parameters and stipulated returns:

The results are hypothetical results and are NOT an indicator of future results and do NOT represent returns that any investor actually attained. Indexes are unmanaged, do not reflect management or trading fees, and one cannot invest directly in an index.

A summary analysis of each study follows. We encourage readers to refer to the corresponding detailed analysis as well as the original research paper as nuance and clarity may be lost in the pursuit of relative brevity.

### The Analysis on Graham’s Value Strategy

### Ben Graham’s Net Current Asset Values: A Performance Update

“Ben Graham’s Net Current Asset Values: A Performance Update” by Henry R. Oppenheimer was published in the Financial Analysts Journal (1986). The study examined the performance of securities that were trading at no more than two-thirds of their NCAV during the 13 year period from 1970-82 period in the US.

### Results and Analysis

Table IV from the study is reproduced below:

Below we have adapted the data from Table IV and looked at the performance in both absolute and relative terms to the S&P 500 Total Return (TR, i.e. *including* dividends):

Disclosure: Performance figures contained herein are hypothetical, unaudited and prepared by Alpha Architect, LLC; hypothetical results are intended for illustrative purposes only. Past ...

more