- To assess whether a stock is cheap or expensive, its valuation needs to be put into context. This is often done by comparing it to its peers.
- An alternative approach is to compare a stock’s valuation to its own history. This “time-series” approach eliminates the impact of inherent differences between companies.
- Combining time-series and cross-sectional approaches can further enhance the analysis of value factors in terms of risk and return behavior.
For value investors, the debate about how best to define the factor and determine cheap vs. expensive rages on. In the previous blog post in our value-factor series, we discussed the effects of accounting for the ever-growing importance of intangible assets. This post explores a way to potentially get a more robust valuation picture by looking to a stock’s peer set and history.
Measuring Valuation
Determining how cheap or expensive a stock is often involves using fundamental ratios such as book to price (B/P) and earnings to price (E/P). These ratios provide useful information, but judging a stock’s valuation requires context. This is often achieved by comparing a stock’s valuation ratios to those of its peers within a specified universe (cross-sectional) or to its own history (time-series).
The selection of the peer universe is important, and the cross-sectional comparison can be a simple ranking or some sort of standardization, such as z-scoring (measuring data points in relation to the mean).1 To eliminate some of the inherent differences between companies’ stocks and across sectors or geography, the standardization is often done within such groupings. For instance, as we discussed in “Bringing Value to the 21st Century,” sectors where companies invest heavily in research and development (R&D) may be deemed expensive, unfairly, because R&D is not considered an intangible asset when calculating B/P.
A time-series approach, on the other hand, is less vulnerable to such differences: They can, to some extent, be mitigated when a stock is compared to its own history.
Comparing Time-Series and Cross-Sectional Approaches
To better understand the impact of time-series standardization on value, we applied it to five valuation ratios — B/P, E/P, cash earnings to price (CE/P), earnings before interest and tax to enterprise value (EBIT/EV) and forward earnings to price (Fwd. E/P).2 We then compared the risk/return characteristics of each with its cross-sectional counterpart. For the time-series approach, the value exposure is calculated by standardizing each stock’s valuation ratio, using z-scoring, within its five-year history of daily data.3 In the cross-sectional approach, the z-scoring is applied cross-sectionally within the appropriate stock universe.
Over the past 20 years, we saw positive returns across all five value descriptors in both the cross-sectional and time-series approaches, although performance varied.
Time-Series and Cross-Sectional Value Performance
Performance of hypothetical pure-value portfolios based on the indicated ratio using the MSCI Global Equity Model for Long-Term Investors (GEMLT). Each long/short portfolio has a unit exposure to its target factor and zero exposure to all other country, sector and style factors. The cross-sectional value factors are the existing ones in GEMLT.
Broken down by decade, we see that performance from 2011 to 2020 — value’s “lost decade” — was lower than in 2001 to 2010 for both approaches (see the exhibit below). Performance for the time-series approach, however, was more consistent across the two decades. This lower volatility enhanced its risk-adjusted returns.
Time-Series and Cross-Sectional Value Performance by Decade
Best of Both Worlds?
Although the time-series and cross-sectional factors are derived from the same data, the correlation of the value-factor exposures, averaged over the 20-year period, suggests the approaches are, indeed, quite different.
Low Correlation Between Time-Series and Cross-Sectional Value Factors
Value exposures using cross-sectional and time-series approaches are calculated on a monthly basis, and then correlation of these exposures across the universe is calculated for each point in time and averaged over the full period.
And there are other differences. As mentioned earlier, the time-series approach is not susceptible to variation across sectors, but it can be affected by changes in the valuation of broad equity markets. For example, a market-wide rise in valuations can cause all stocks to rise to extreme valuation levels over the five-year period, and would not lead to meaningful differentiation of valuations. On the other side, by not relying on historical data, the cross-sectional approach is able to quickly capture any changes, but also tends to be less stable.
Given their low correlation and complementary characteristics, a combined approach may provide a more comprehensive valuation measure. To assess this, we look again at the behavior of each valuation factor, this time calculated by averaging cross-sectional and time-series scores.
For the full 20-year period, the combined versions demonstrated higher absolute returns than each individual approach, but greater volatility. Risk-adjusted returns were lower for the combined version versus the time-series approach for all factors except B/P, but were higher than those of the cross-sectional approach across all factors.
Focusing on the decades individually, we saw that some of the improvement in the time-series approach from 2011 to 2020 faded when combined with the cross-sectional approach. From 2001 to 2010, however, the combined approach performed in-line with or better than either approach across all value factors.
Risk/Return Characteristics of a Combined Approach
Data from 2001 to 2020
One Plus One May Equal More than Two
Whether constructing a fundamental factor model, a value strategy or a value index, valuation ratios need context. While time-series and cross-sectional approaches each have pros and cons, their complementary behaviors and relatively low correlations suggest investors may have been able to get a clearer valuation picture by combining the two.
1Standardization is usually done using a z-score: zi = (xi – μ) / σ, where xi is the raw data (e.g., P/E ratio) for stock i and μ and σ are the average and standard deviation across the universe.
2These valuation ratios are the basis for the construction of two value factors in most MSCI equity factor models, such as GEMLT.
3Our results were not significantly sensitive to the period used. We select five years to have enough data points to make the comparison meaningful while avoiding long-term broad shifts in valuations.
Further Reading
Bringing Value to the 21st Century
Did Value-Factor Exposure Deliver for Value Funds?
MSCI Perspectives podcast: In Spring, Does an Investor’s Fancy Lightly Turn to Value?