When valuation fails

Introduction

As an industry, we like to publish research that “works” (at least on paper). Otherwise, we sweep failed experiments under the rug. This problem is not specific to finance. Academics in all fields are in a never-ending quest for high r-squareds and t-statistics.

This article is different. In the context of the 50th anniversary issue of this journal, we hope it will encourage researchers to contribute more articles on what doesn’t work as much as what does. We learn more from mistakes, failures, losses, and setbacks than when everything goes as expected. “The obstacle is the way,” wrote modern philosopher Ryan Holiday¹.

We examine the failure of a basic model: the relative valuation discipline at the core of most tactical asset allocation (TAA) processes. Nowhere has this been more evident than in the failure of value stocks to catch up to their growth counterparts since the Global Financial Crisis of 2008–2009. We attribute this failure to technology disruption and accounting’s misreporting of intangibles. We discuss how skilled asset allocators have avoided this value trap’s gravitational pull, including with a momentum adjustment, which has been one of the simplest ways to cure the relative valuation malaise. When an asset class is cheap and shows positive short-term momentum, the valuation signal works better².

We also propose a new methodology to evaluate backtesting performance, that is, what works. Statistical measures, with their beautiful star notation (**, ***), project rigor and appease referees. However, they are silent on the sensitivity of the model’s performance to the choices of lookback data windows and portfolio construction (“bucketing”) methodologies. Our data mining confidence bands complement traditional statistical measures to provide transparency on model parameter choice. This information is what practitioners need to know if they want to avoid using overfitted models.

How is “value” defined and is there a value premium?

An easy way to distinguish between value and growth stocks is to rank them by their book-to-market ratio (B/M). This ratio can be a yardstick for disagreements between accountants and money managers about a company’s value.

• Book value (B) is what accountants think the company is worth—the difference between the company’s assets and liabilities.
  
• Market value (M) is the market capitalization (shares outstanding multiplied by price). It’s what investors think the company is worth based on its future earnings prospects.

Theoretically, companies with a high B/M are “cheap,” and those with a low B/M are “expensive.” Of course, it’s not that simple. Valuing a company is difficult. It requires forecasting future earnings and putting a price on risk. On the one side, accountants use rules. They try to avoid making messy judgment calls. These rules are necessary to make financial statements comparable across companies but often fail to capture future earnings growth. Despite efforts to measure “intangibles,” accountants don’t have the tools to value a fast-growing company’s ability to gain market share. There’s too much judgment involved.

On the other side, it’s an investor’s job to make judgment calls. We don’t care whether our forecasts are comparable with those of other investors—active managers want a proprietary edge. Two accountants should not disagree on a company’s earnings or book value, but money managers are expected to disagree on a company’s market value. A stock’s price reflects a collection of independent judgments.

Over time, who’s been right more often? Accountants or money managers? There’s evidence that money managers should pay more attention to book values.

Between 1926 and 2023, stocks with high B/M outperformed those with low B/M by an average of 4.2%². Academic careers have been built on discovering and explaining this value premium, and money management careers have been built on harvesting it.

Academics explain it as compensation for risk. Value stocks are more cyclical; hence, investors should require a premium to invest in them (see, for example, Fama and French 1992 and Zhang 2005). Some money managers prefer to explain it as an anomaly caused by irrational investor behavior. They posit that value stocks have outperformed over time because they’re boring. The idea is that investors tend to overpay for “glamour,” high-growth, and high-momentum stocks (Hagens and Magwa 2022).

Unfortunately for the value zealots, the value premium has weakened. The average value premium over the last 20 years was −1.4%. Growth stocks outperformed despite their low B/M³.

The TAA Problem

This disappearance of the value premium has challenged not just stock pickers but also tactical asset allocators, most of whom use relative valuation metrics to overweight cheap and underweight expensive asset classes. This strategy sounds simple, but it’s not. There are two big challenges.

• It isn’t easy to catch turning points. To unlock a valuation advantage, you need a catalyst. That’s why most TAA processes incorporate fundamental, macroeconomic, and sentiment factors.
  
• Secular changes can create value traps. For the last 20 years, the relative valuation of value stocks has trended down, as shown in Exhibit 1. Relative to growth stocks, value stocks have gotten cheaper and cheaper… and cheaper. For investors who seek to make money from relative valuations reverting to the mean—which historically has tended to work over time and across asset class pairs (Page 2020)—that’s a disheartening chart.

Our study evaluates the implications of this lack of mean reversion for TAA models.

¹See Holiday (2014).
²See Asness, Moskowitz, and Pedersen (2013) and Bhansali et al. (2015).
³The premiums of 4.2% and −1.4% are from Kenneth French’s data library: 4.2% is the arithmetic average of annual (calendar) “HML” (high minus low B/M) factor returns between 1927 and 2023, and −1.4% is the average for the last 20 years (2004–2023).

Tactically overweighting value or growth based on their relative valuation delivered modest excess return over a 50%/50% value/growth benchmark, as shown in Exhibits 2 and A5. From August 1936 to March 2024, the average after-cost cumulative alpha of the 24 valuation-based trading rules was 19% or 20 bps per annum. The data mining confidence band was (−22.1%, 58.9%). A quarter of the strategies underperformed the benchmark. Interestingly, most of these losing strategies are characterized by a long lookback horizon (60 or 120 months). Hence, the first lesson is to avoid comparing current valuations with the distant past. This is unsurprising given the tectonic shifts of the value and growth styles discussed earlier.

The average information ratio of the 24 trading strategies, computed on the full August 1936–March 2024 sample, was 0.06. Its data mining confidence band was (−0.03, 0.15).

In Exhibits 3 and A6, we show the information ratio over rolling 20-year periods. Over the last 20 years, the average information ratio across the 24 strategies was −0.01. This is a clear indication that the relative valuation signal’s effectiveness disappeared in the last two decades.

Valuation has failed as a stock selection approach and tactical allocation signal. However, for those who like to look at the glass half full, the current 20-year weakness is not unprecedented. The information ratio followed a cyclical pattern. In the late 1950s and during the 1980s, it was at levels comparable to the present. It recovered in the 1960s–1970s and in the 1990s.

In any case, buying cheap is not enough: We need a catalyst that reflects the trend toward repricing. Consistent with Asness, Moskowitz, and Pedersen (2013), momen-tum improves the performance of valuation-based strategies. When we control for momentum, the average cumulative alpha across the 24 valuation-based strategies improved from 19% to 48.8%, as shown in Exhibit 4.

Of the 10 alternative momentum definitions, only one generated a lower average cumulative alpha than the one based on relative valuations only. We obtained the best outcomes by measuring momentum in terms of trailing 6- and 12-month relative returns and the difference between the 12-month and 36-month relative returns. The worst outcome occurred when we defined momentum based on the rolling 36-month relative return. This indicates that such a horizon is too long to capture momentum.

Across all the strategies, when we combine valuation and momentum, the average cumulative alpha increased from 19%, with a (−22.1%, 58.9%) data mining confidence band, for the relative valuation signal to 48.8%, with a (3.9%, 101.8%) data mining confidence band. Momentum delivered a risk-adjusted performance improvement in terms of information ratio, as shown in Exhibit 5. When we combine valuation and momentum, the full-sample information ratio increased from 0.06, with a (−0.03, 0.15) data mining confidence band, for the relative valuation signal to 0.13, with a (0.02, 0.26) data mining confidence band.

Momentum helps avoid value traps. Nonetheless, the information ratio over the last 20 years was low and close to the bottom of the historical range, even if we control for momentum. Better capturing inflection points enhances the relative valuation signal’s performance, but it is not a silver bullet. It doesn’t amend the structural headwinds value stocks have experienced.

Exhibits A3 and A4 in the appendix report cumulative alphas and full-sample information ratios for the 24 valuation-based trading strategies, either with no momentum control or combined with each of the 10 alternative momentum definitions. In total, we have 264 performance measures.

Discussion

What has created the mother of all value traps over the last 20 years? In one word, technology. Lev and Srivastava (2022), explain that corporate business models have shifted from investing in hard assets (property, plant, and equipment) to intangibles (research and development). The breakthroughs from these intangible investments have been highly disruptive to legacy business models. Unfortunately, accounting practices have failed to adapt to this shift. According to the authors, growth companies—especially tech companies—that invest in intangibles have looked increasingly expensive due to deflated book values. “A firm investing heavily in R&D, IT, brands, or business processes (e.g., customer recommendation algorithms), may appear to be an overvalued company … whereas in reality its valuation isn’t excessively high when book value is properly measured,” they explain (Lev and Srivastava 2022).

Earnings and cash flows have also been understated because intangibles are immediately expensed. This has led to inflated price-to-earnings ratio (P/E) and price-to-cash flow ratio (P/CF) for growth stocks. Therefore, over the last 20 years, using P/E or P/CF to construct value portfolios did not perform better than using P/B. The premium was −2.1% when using P/E and −2.3% when using P/CF, compared with −1.4% for P/B, as mentioned earlier⁶.

To be clear, it’s not just the fault of accountants. Like relative valuations, fundamentals have trended rather than mean-reverted. As measured by return-on-equity (ROE), profitability has steadily improved for growth relative to value stocks, as shown in Exhibit 6. Value has underperformed because platform companies operating in a digital world have demonstrated sustained and exceptional growth.

⁶The premiums of −2.1% and −2.3% are from Kenneth French’s data library: −2.1% is the arithmetic average of annual (calendar) high (top tercile) E/P stocks’ value-weighted returns minus low (bottom tercile) E/P stocks’ value-weighted returns between 2004 and 2023. The −2.3% value is the arithmetic average of annual (calendar) high (top tercile) CF/P stocks’ value-weighted returns minus low (bottom tercile) CF/P stocks’ value-weighted returns between 2004 and 2023.