What Has Worked in Investing – Examining Net Nets
Celebrated value investors Tweedy, Browne Company LLC published a booklet in 2009 that contained over 50 studies titled “What Has Worked in Investing: Studies of Investment Approaches and Characteristics Associated with Exceptional Returns”. Of particular interest to us is their inclusion of a study on the returns achieved by “net nets”.
The objective of the original study was to examine the performance of securities that were selling at 66% or less of net current asset value from April 1970 to April 1981, thereby creating a 12 year study period.
Our objective was to analyze the study itself; determine its reliability, draw our own conclusions and glean, if any, actionable advice for the practitioner of the “net net” method of investing.
While a specific formula was not mentioned on page 5 where the study commences, earlier in the publication the authors mention “The net current assets investment selection criterion calls for the purchase of stocks which are priced at 66% or less of a company's underlying current assets (cash, receivables and inventory) net of all liabilities and claims senior to a company’s common stock (current liabilities, long-term debt, preferred stock, unfunded pension liabilities).” (p.1)
We therefore assume this is the formula applied, with the exception of the incorporation of “unfunded pension liabilities” which would have required a review of the annual reports for each company to identify – a task unlikely to have been undertaken for the purposes of this study.
Weighting: Not specified
Purchase/rebalance date: “April 30 in each of 1970 through 1981”
While rebalancing on the same date each year is consistent it nonetheless exposes the results to “timing luck”. (To limit the impact of timing, researchers often create portfolios each month using overlapping periods).
Holding/rebalancing period: “investment returns for all stocks were computed for 6 months, 1 year, 2 years and 3 years after each selection date.”
While there are numerous biases/errors that can be made when conducting studies/back tests, below we have analysed those we deem most likely to impact a study of this nature:
1. Survivorship bias
“All 7,000 public companies in the Compustat database, including the Research File of companies which had been acquired, merged or declared bankrupt subsequent to an assumed historical selection date”
The data source appears to be free from “survivorship bias”.
2. Look Ahead bias
Controlling for “look ahead” bias was not specially mentioned, however, given the rebalance date of 30 April it is likely that the study used financial statements ending 31 December, thereby avoiding look ahead bias.
3. Time period bias
The study spans 12 years and we classify this as a “somewhat reliable” period.
· < 10 years; inadequate/unreliable
· 11 to 20 years; somewhat reliable
· > 20 years; more reliable
· > 40 years; most reliable
4. Data source and treatment
“All 7,000 public companies in the Compustat database”.
Compustat is a reputable database.
It should be noted that the study included firms with a “market capitalization of at least $1 million” i.e. “nanocap” firms formed part of the investment universe examined. However, no mention of the number of firms that qualified as investment candidates was mentioned. We understand that the Compustat database includes securities listed on the NYSE, AMEX and NASDAQ exchanges.
Interestingly, the Oppenheimer study was conducted over a nearly identical time period and he did specify the number of securities in his study.
Table I from Benjamin Graham’s Net Current Asset Values: A Performance Update by Henry R. Oppenheimer is reproduced below:
As Compustat does not include OTC securities we note that for some years the sample of stocks listed on the NYSE and AMEX dwindled. With regard to the Tweedy Browne study however, presumably, the NASDAQ exchange would have presented other net net opportunities thereby increasing the sample size studied.
5. Human error
There is nothing specified in the research methodology that would make us believe this study is at greater risk of suffering from human error.
6. Journal rating/credibility[i]
This study was not, to our knowledge, published in a top tier academic journal and therefore cannot be granted the “additional credibility” that may come with such publication.
Reliability Assessment: Aside from the relatively short time period, 12 years, the study appears to be reliable.
Results and Analysis
Table 4 from What Has Worked in Investing is reproduced below:
The “Average Return” is almost certainly the arithmetic mean return which overstates the actual returns attainable by an investor in practice. We reached this conclusion as a number of other studies published in the booklet expressly state that returns were “compounded”, whereas in this case no such specification was made. Furthermore, using historical data we found that the specified 1 Year S&P500 of 8.5% reconciles almost perfectly to the S&P500 total return arithmetic mean return calculated using calendar year returns[ii]. Oddly, the 1 Year S&P500 returns presented against “Stock Selection Criteria” “66% of net current asset value” is different to all other portfolios (i.e. 9.1% vs 8.5%). Presumably this is because a different time period was used for these portfolios, but we did not note any explanation for this difference in the study.
Further, it is stated that “One million dollars invested on April 30, 1970 and rolled over at each subsequent April 30 into the stocks selling at less than 30% of book value would have increased to $23,298,000 on April 30, 1982. One million dollars invested in the S&P 500 on April 30, 1970 would have been worth $2,662,000 on April 30, 1982.” i.e. $1m compounded at 30.0% for 12 years.
An otherwise astute reader may take the stipulated ending value of $23,298,000, starting capital of $1,000,000 and the 12 year time horizon to “confirm” the apparent geometric mean of 30.0% (i.e. r = (23,298,000/1,000,000)^(1/12) -1). While seemingly astute it is nonetheless inconsistent with what was actually reported – the arithmetic mean. By ostensibly attempting to be helpful and express the returns in dollar terms, the authors may have inadvertently delivered something of a “one-two knockout” to the reader; one, use a method that overstates returns (arithmetic mean) and two, utilise an example implying the result represents the compound annual growth rate to further exacerbate the issue!
This is another reminder to actively go in search of disconfirming evidence and question everything, irrespective of its source.
“The important thing is not to stop questioning.” (Albert Einstein)
Now, for all our concerns over the methodology used to calculate returns let’s revisit the geometric mean return reported in “Benjamin Graham’s Net Current Asset Values: A Performance Update” which, as previously mentioned, was conducted over a nearly identical time period:
Interestingly, the geometric mean reported by Oppenheimer of 28.2%, is only marginally lower than the 28.8% “Average Return” stipulated for portfolios of net nets held for 1 year. With the 1973-1974 bear market this lack of divergence between the arithmetic and geometric mean seems curiously small notwithstanding that different exchanges were examined. (Please note, we discussed the potential divergence between the geometric and arithmetic mean with specific reference to volatility in “Testing Benjamin Graham’s Net Current Asset Value Strategy in London”.)
Despite that identified above, we would always caution against putting outcome ahead of process.
What Has Worked in Investing is a wonderful resource; nonetheless it is important to read each study contained therein analytically and be cognisant of the different methodologies used in calculating and presenting the reported returns.
While in many ways this study is satisfactory, for a number of reasons, we would not place any great reliance on it. Firstly, it covered only 12 years, a relatively short period when trying to ascertain the efficacy of an investment strategy. Secondly, it invested in companies with market capitalizations as low as one million USD, which, even by today's inflation adjusted standards would likely see many investors face liquidity constraints and consequently a “limit to arbitrage”. Most critically, however, the “Average Return” reported in the study was calculated as the arithmetic mean of returns, which would have overstated the actual returns achievable by an investor as measured by the more appropriate geometric mean.