How Accounting Is Making Value Investing Harder
Introduction
My prior post (and the OSAM article that inspired it) described the declining usefulness of book values to investors. The OSAM article gives some reasons for the deterioration of book values: the rise of intangibles, which are not recorded in book value, along with increasing use of accelerated depreciation and buybacks, both of which decrease book values. However, that is not the end of the story.
As Baruch Lev pointed out in his book “The End of Accounting” (2016), most issues that affect book values also affects earnings since double-entry bookkeeping requires that the change in book value (minus dividends) is equal to earnings. Hence, if book values are incorrect due to missing changes in intangible assets, then earnings are incorrect as well.
Indeed, while the deterioration of P/E outperformance has not been as stark as that of P/B, which has hugely underperformed since 2003, it is still present: the cheapest quintile of large cap stocks by P/E slightly underperformed the most expensive quintile over that same period.
The academic literature suggests that this trend is only accelerating. In his book, Lev shows that the ability of earnings to predict market values has decreased drastically since the late 1970s. Likewise, Anup Srivastava, in his paper “Why have measures of earnings quality changed over time?” (2014), found that the problems are increasingly worse in younger firms, which have a higher use of intangibles, leading to earnings that are more volatile and less relevant for predicting both market values and future earnings. Hence, the underlying trends suggest the problem will get worse over time not better.
Why is this happening? And why did the problem start specifically in the 1970s? In this article, we’ll try to find the answer. In order to do so, we need to dive into some details of accounting.
The Matching Principle
A primary benefit of accrual accounting over cash accounting is the ability to match costs with the revenues that they generated, i.e., to record the costs and revenues in the same period so that financial reports show the proper return on investment. In 1953, A.C. Littleton called matching the “central purpose of accounting”.
To see what can go wrong without matching, let’s look at an example. Suppose that we are running a farm (though this could just as well be a software company, a movie studio, or many other types of businesses). In this setting, we realize costs of planting seeds in the spring, and we generate revenues by harvesting the crops in the fall. If we created our farm by investing $100k in 2019, then our financials in 2020 might look like this:
Planting seeds in the spring had an up-front cost of $100k but harvesting in the fall generated revenues of $110k, generating a total profit of $110k – $100k = $10k and a return on (prior year) equity of $10k/$100k = 10%.
That all looks fine. But now suppose that our farm was on an alien planet where seeds were planted in the fall and harvested in the spring. Now, our accounts in 2020 would look quite different:
In 2020, since we have recorded our planting expenses but have not yet seen our harvesting revenues, we show a loss of $100k. In 2021, we record our revenues and show a profit of $110k. Without proper matching, the two years completely misrepresent the function of the business. It appears that lost all of our shareholder’s money in 2020 and then miraculously made $110k in 2021 starting from no money at all (an infinite return on equity!).
You might suggest that we can solve the problem by looking back two years to 2019. That would work in this example, but the example is easily changed. Suppose that our farm continues operating for another two years (as it should). If we reinvest all the money we made each year, then we will appear to have a money losing operation in those years as well:
From 2020–22, we made no profit. Only once we finally stop reinvesting the funds in 2023, do we finally see that the company was earning a 10% return on equity, compounding each year, during the whole period.
This last example may remind readers of Amazon, which continued to record no profits (and pay no taxes) for years, despite an underlying business that was incredibly profitable.
Also, note that the problems described above would not be fixed by using cash accounting rather than accrual accounting. This was cash accounting! The problems above were due to poor matching between the time periods of the costs and the revenues. Accrual accounting is what would allow us to fix this.
To do so, we need to avoid realize the costs of planting seeds in 2019. We do that by creating an asset for the seeds in the ground, recorded at cost. The creation of that asset in 2020 creates a positive flow of $100k, which negates the cost of planting the seeds. We can depreciate that “seeds in the ground” asset in 2021 for a negative $100k. The changes (shown below in red) allow us to move the expenses from 2020 to 2021, when the harvesting occurs:
These properly matched financial reports show the business results more clearly. In 2020, before any harvesting, we had no change in book value and no returns, but in 2021, after harvesting, we see a return of 10% from the prior year’s book value.
All of this is theory, so a natural question to ask is whether this is really happening in practice. The answer is sadly yes. Above, we already mentioned the work of Srivastava. He found that younger firms, who have more focus on intangibles, also had poorer matching. In his article, he used the technique introduced by Dichev et al. in their paper “Matching and the changing properties of accounting earnings over the last 40 years” (2008).
Dichev et al. studied matching directly by using linear regression to predict each firm’s revenues from expenses in the current, prior, and next years. Prior to the late 1970s, the prior and subsequent years had no predictive power. The current year expenses contained all the predictive power for that year’s revenues, suggesting good matching. Since then, however, both prior and next year expenses have become relevant in predicting current year expenses, suggesting poor matching. Furthermore, the effect was increasing over time.
Once again this specific timer period, late 1970s, arises in the data. What happened then that would affect the quality of matching?
Two Schools of Thought
Above, I quoted A.C. Littleton, the author of the preeminent accounting textbook of the period, who said, in 1953, that the matching was the central purpose of accounting. However, that is not the only view of accounting and its purpose. Matching is the central purpose for those who view the income statement as preeminent. To them, the balance sheet serving an important but secondary role in recording costs that await corresponding revenues. However, a different view takes the balance sheet as preeminent, with income arising simply as the change in book value. To them, the core purpose is correctly stating the value of assets and liabilities.
There is an inherent tension between these two viewpoints. In our example above, we included “seeds in the ground” as an asset valued at $100k. But are they really worth that amount? It doesn’t matter as far as matching is concerned, but if you want the balance sheet to correctly state the value of shareholder equity, then you could be concerned about this new asset, whose value is certainly questionable.
As Fera et al. document in their article “A Renewed Interest in the Fundamentals of Accounting: The Impact of the Matching ‘Principle’ on Earnings Attributes”, due to this tension, until the 1970s, accounting was a compromise between the income statement and balance sheet views. However, in 1973, FASB became the standard setter for accounting and decided that a “muddled” compromise was unwise, choosing the balance sheet approach as the “conceptually superior” method of accounting. By the late 1970s, the balance sheet view was entrenched in accounting standards.
As we just saw, this reduces the priority of matching. Our seeds in the ground example was hypothetical, but realistic examples abound. The true value of R&D spending is not easy to determine, making it too suspicious to be recorded as an asset. Instead, it is immediately expensed. One could make a similar argument for the costs of shooting a movie, advancing a drug through Phase I, II, or III trials, or acquiring a new Netflix subscriber. Those would be assets for the purposes of matching, but their true “values” are hard to pin down, so they do not belong on a pristine balance sheet.
From my view as an investor — and notably not an accountant — the balance sheet view feels like it is trying to do my job (determining the value of the company) for me. And frankly, it is doing that job poorly. Proper matching, which allows me to determine return on investment, is the information I need to determine the value myself, and with that information, I am happy to do so. A mismatched set of costs and revenues makes determination of return on investment very difficult, regardless of how confident I can be about the values of assets and liabilities. (In a liquidation, perhaps, the latter would be more useful, but it is not so for valuing a going concern.)
Conclusions
As I noted above, I am not accountant, so everything I said above should be taken with a grain of salt. There certainly could be important facts that I am missing. However, from where I sit, there is not only a strong correlation between the declining quality of matching in earnings and the declining usefulness of earnings in predicting market values but also a good causal explanation for why the former would cause the latter. The decline in the usefulness of earnings appears to be a result of FASB’s decision to favor the accuracy of balance sheets of over the accuracy of earnings.
As Srivastava’s work demonstrates, these problems are worse in younger firms, which rely more on intangible assets. As those firms take up more of the total market capitalization, the declining usefulness of earnings will become a problem that few investors can avoid.
For now, when the provided financial reports have poor matching, properly valuing a business requires a significant amount of investigation.
For one example, see (or rather, listen to) Ron Baron’s description of the unit economics of each Tesla car factory. Assuming Baron is correct, each factory is an extremely profitable investment, yet Tesla itself reported no earnings until 2020 (and even then, the profits were mainly from buying bitcoin!).
For another example, see this Business Breakdown of Chipotle, which discussed the unit economics of each store. Once again, determining the true return on investment for an individual store is difficult if not impossible from the poorly matched income statement. When costs are expensed before revenues, overall profitability is obscured. Furthermore, the faster the company is growing, the less profitable it appears, despite being more valuable in reality.
If these examples are representative, then investors will need to become diligent investigators in order to determine the the true value of young, growing businesses because standard financial reports do not provide us with the information required to do so.
Finally, let me mention Ilia Dichev’s article “On the Balance Sheet-Based Method of Financial Reporting” (2007), which not only provides a good overview of the history of financial reporting rules but also provides a useful idea for how to fix these problems. He suggests using the income statement view for operating assets and the balance sheet view for non-operating assets. The former, which accurately represent both how managers and investors think about the businesses, presents an accurate income statement albeit with a balance sheet that is mostly “unexpired costs” (like we saw for our farm above). The latter, which accurate represents the assets that could easily be separated from the business and sold, presents an accurate assessment of their worth, while any earnings that result as changes in their values are largely noise that should be ignored by investors. This approach allows both operating and non-operating assets to be most accurately valued, with the entire enterprise value produced as the sum of those parts.
