“How can a q-theoretic model price momentum?” is a new paper by Robert Novy-Marx and goes right to the heart of an intense debate ongoing in empirical asset pricing — can neoclassic economic models explain the so-called momentum anomaly? A quote from the start of the paper, which answers the question of “Can the investment CAPM price momentum?”
“The answer, of course, is that it can’t” — Sentence one of the abstract. 1So a reader quickly gets the answer! This paper questions some of the findings in Lu Zhang’s Replicating Anomalies paper. For those unfamiliar with Lu’z research, we have highlighted the replicating anomalies paper here and Wes has a long interview with Lu here (as well as an interview on a podcast). High-level, Lu’s Investment CAPM (also know as the q-factor model) says that the cross-section of expected returns should be explained by two main factors, (1) Investment and (2) Expected Profitability. 2 Lu adds the market and size as the third and fourth factors when empirically examining the cross-section of returns. In the Replicating Anomalies paper, Lu mentions that his model with his four factors can explain almost all of the anomalies! And the list is long, as they test 447 anomalies! A few of the factors that they can explain are (1) Gross-Profitability and (2) Momentum (both Price and Fundamental Momentum). Enter Rober Novy-Marx. Why? Well, here are some of his papers:
- The Other Side of Value: The Gross Profitability Premium. (JFE 2013)
- Fundamentally, Momentum is Fundamental Momentum (working paper, described on our site here)
The Paper Set-up and ResultsThe paper begins by highlighting two of the main results in the Replicating Anomalies paper:
- The “expected profitability” factor (as measured by ROE) yields significant excess returns, even as measured by the 3 and 4-factor models.
- The q-factor model prices Price Momentum (UMD).
In other words, we get the following:
So the ROE variable in Lu’s paper has been decomposed into two variables–(1) Lagged ROE and (2) Change in Earnings scaled by Book equity. The first variable is fairly straightforward, one-year lagged earnings over lagged book equity (lagged ROE). The second variable can be thought of as either (1) change in earnings scaled by book equity, or (2) earnings innovations scaled by book equity or (3) fundamental momentum. So a natural question arises–of the two variables (lagged ROE and Earnings Innovations)–which one is most responsible for the excess returns? Or do both contribute to the excess returns? Naturally, this is tested in the paper. Table 2 highlights the results to five long/short portfolios (from a set of 25 portfolios), created by varying the initial sorting variable. Panel A first sorts firms into quintiles on earnings innovations, and then re-sorts on lagged ROE. The five long/short portfolios are formed by going long the high lagged ROE quintile and going short the lagged ROE quintile, within each Earnings Innovations quintile. 7 Panel B reverses the order of the sorts–first sorting by lagged ROE and then sorting on Earnings Innovations. So which variable is more important? The results are shown below in Table 2: Panel A finds that the average excess returns are insignificant (as can be seen through the 4-factor alphas)–thus the lagged ROE does not appear to generate significant abnormal returns. However, Panel B finds significant excess returns–thus it is the Earnings Innovations that are generating excess returns. So when one decomposes ROE into two variables, lagged ROE and Earnings Innovations (i.e. fundamental momentum), one finds that the significant excess returns appear to be driven by the fundamental momentum/Earnings Innovations. To further bring home this point, Robert examines the long/short returns to five strategies:
- PEAD — Post Earnings Announcement Drift. Monthly-rebalanced portfolios on their SUE 8
- Delta ROE — The Earnings Innovation/Fundamental Momentum described above
- ROE — The ROE variable used in the q-factor model
- E/B — Annually rebalanced Earnings/Book Value of Equity
- lag ROE — Lagged ROE as described above
- Regression 5 generates the same results in the Replicating Anomalies paper, which is that the ROE factor can price the UMD factor (price momentum).
- Regression 6 finds that fundamental momentum (PEAD) also prices the UMD factor.
- Regression 7 finds that adding the ROE factor alongside PEAD only yields marginal ability to explain return variations (as seen by the R^2 barely increasing).
- Regression 8 finds that fundamental momentum (as measured by the Earnings Innovation factor) also prices the UMD factor.
- Regression 9 finds that adding the ROE factor alongside Earnings Innovations only no marginal ability to explain return variations (as seen by the R^2 not changing at all).
ConclusionSo the results above call into question the ability of the q-factor model to price two variables (momentum and profitability) common in the standard 6-factor model (the Fama and French 5-factor model + momentum). By simply decomposing ROE into its two components–lagged ROE and fundamental momentum (earnings innovations)–one finds different results than in the Replicating Anomalies paper. However, at the end of the day, if we pretend the q-factor model has 5 factors (not 4), by splitting ROE into (1) lagged profitability (i.e. lagged ROE) and (2) earnings innovations (i.e. fundamental momentum)–we would get similar results, in that both profitability and momentum (either measured by earnings or price) are important! Additionally, it should be noted that in Lu’s new working paper, “q^5”, they embed the Earnings Innovation/Fundamental Momentum variable into the 5th factor (this is “dROE” in the “q^5” paper). 12 An explanation of the new paper and factor can be found here. So what is the big-picture takeaway? While it may appear that the two authors are competing, in my opinion, they have found similar results, but have different economic interpretations of those results. I don’t think that these findings discredit the Investment CAPM. However, these findings highlight that embedded within the ROE variable is a “fundamental momentum” variable, which appears important to pricing momentum in factor regressions. 13 As we have mentioned before, and as the competing results above show, factor investing is more art and less science. Let us know what you think …
How can a q-theoretic model price momentum?
- Robert Novy-Marx
- A version of the paper can be found here.
- Want a summary of academic papers with alpha? Check out our Academic Research Recap Category.
The answer, of course, is that it can’t. Hou, Xue, and Zhang’s (2014) empirical model does price portfolios sorted on prior year’s performance, but for reasons outside of q-theory—it does so by including a fundamental momentum factor, i.e., a factor based on momentum in firm fundamentals. The ROE factor, which does all the work pricing momentum, is constructed by sorting stocks on the most recently announced quarterly earnings, which tend to be high after positive earnings surprises. A post earnings announcement drift factor prices the model’s ROE factor, and subsumes the role the ROE factor plays pricing momentum portfolios when both are included as explanatory variables. The HXZ model also only prices portfolios sorted on gross profitability by conflating earnings profitability, which drives the ROE factor’s covariance with gross profitability, with post earnings announcement drift, which drives the ROE factor’s high average returns. Controlling for fundamental momentum, the HXZ model also loses its power to explain the performance of gross profitability. These facts are inconsistent with a neoclassical interpretation of the empirical model.
- I took the liberty of describing the “q-factor” model as “Investment CAPM”. ↩
- Here is a link to where Lu outlines the Investment CAPM from the start ↩
- For those interested in other academic wars we have covered, here is (1) AQR versus the academics and (2) Trading costs wars ↩
- What does it mean to say that a model or a variable “prices another variable? Well, imagine that we have two variables, named variable A and variable B. When we run a baseline regression (regressing variable A and variable B against a model), we see that both have “excess” returns (after controlling for other factors, such as the market, Value, Size, etc.). If we then regress Variable A against a model (such as the 3-factor model) and also add Variable B as a dependent variable (to create a 4-factor model), if the addition of Variable B in the model causes the excess return (alpha) to disappear, we would say that Variable B “prices” Variable A. ↩
- I asked Lu what is the economic reasoning behind using a monthly rebalanced variable to proxy for expected profitability. Here is his response: “Using monthly updated Roe is easy to justify. Suppose profitability follows an AR(1) process – a fairly innocuous assumption that is well established empirically, e.g., Fama and French (2000, Journal of Business). It follows that the latest known Roe (including recent earnings innovations) contains better information about future Roe than 4-quarter-lagged Roe.” ↩
- Income before extraordinary items ↩
- So long/short portfolios are formed on the second sorting variable. ↩
- Standardized Unexpected Earnings. Here is the definition from the paper: “Standardized unexpected earnings (SUE) is calculated as the most recent year-over-year change in earnings per share, scaled by the standard deviation of the earnings innovations over the last eight announcements, subject to a requirement of at least six observed announcements over the two-year window. For earnings per share, I use Compustat quarterly data item EPSPXQ (Earnings Per Share (Basic) / Excluding Extraordinary Items). Earnings announcement dates are Compustat quarterly data item RDQ. ↩
- It is unclear if the other strategies were levered up or de-levered to achieve the same volatility. ↩
- I think both are accurately described. The first variable, earnings innovations, is explicitly examining the changes in earnings and scaled by book value of equity–one can view this as fundamental momentum (i.e. earnings momentum). The second variable, SUE, measures changes in earnings over the last year and scaled by the volatility of earnings–again, this is one way to measure earning (fundamental) momentum. ↩
- This is true with or without lagged ROE. ↩
- Note–this is not the exact 5th factor, but is used to create the 5th factor ↩
- For non-factor regressions, one finds a different result. The structural estimation shows that the expected investment growth is more important than the expected profitability in explaining momentum. See the detailed discussion on p. 20-22 in Goncalves, Xue, and Zhang (2017) — Thanks to Lu for pointing this out. ↩