This is the fourth post in a series discussing some of the problems associated with investing in “smart beta” strategies. For the previous post on the how factors can change, click here.
There’s an old story about two economists who were walking down the street, when one stops and says, “Look, there’s a £20 note on the ground!”
The other turns and replies, “Can’t be. If there was a £20 note on the ground, somebody would have already picked it up.”
This story is a good way to think about both sides of the efficient market hypothesis. One side, the inefficient market believers, thinks there can be a £20 note there, and it’s worth the time and effort to search for more of them. The other side, the efficient market believers, believe that there can’t really be a £20 note there – or at least it’s so unlikely that it’s foolish to spend time searching for more of them. If there was a profitable system for picking £20s off the floor, someone would have found it already.
The more people that know about a market inefficiency – or the more people that know that £20 notes are being dropped on the floor – the more likely it is that people will try and correct the inefficiency – the more people will look for the £20 notes. The inefficiency corrects itself.
The same may apply to factor investing. In theory, the more people that know about the excess returns on offer from factor investing, the more people will try and exploit it, and so the lower the expected excess returns become. As more investors allocate more to a known factor, the factor’s abnormal returns start to evaporate as the pricing inefficiency is correct by market participants.
This situation does seem to be reflected in the reality of academic evidence, at least in the US. There’s a growing body of research showing how factor returns decline after research is published and the factor becomes widely known, although the findings are less conclusive for international investors.
If we start our journey in the US, researchers found that when studying 97 different possible factors, portfolio returns were 58% lower after the results of the factor’s effectiveness were published in academic journals. Post-publication declines were also found to be greater for those factors that had highest in-sample returns. (In-sample returns are those found in academic backtests, out-of-sample returns are those found in real-life, following publication). So those factors that looked the best in backtests, suffered the largest drop-offs in returns after publication.
Source: McLean and Pontiff
These findings are backed up by research from Research Affiliates, who suggest there at least three drivers likely contribute to this return deterioration:
1) if researchers try many different definitions of predictors, many factors may show high in-sample returns purely by luck or may materially overstate the true factor potential (some may even be entirely spurious)
2) after discovery, many investors try to exploit the anomaly so that the returns weaken and trading costs rise as the trade becomes crowded
3) the backtest returns could simply be a result of correlation with other factor exposures (e.g., a factor might have been “cheap” at the beginning of the backtest and “expensive” at the end in the context of value).
Source: Research Affiliates
This fantastic chart shows how that the actual post-publication returns from a portfolio of 46 factors (black line), haven’t been as strong as was predicted by how the factors performed pre-publication (red line). After the factors became known, the actual returns were much lower than the research would’ve predicted.
As most of the major factors had been discovered by 2003 (with the exception of operating profitability), the authors also show how those factors have performed since 2003, to give an idea of returns since the factors had been discovered. They found that other than the market factor, not a single factor had delivered a statistically significant excess return since 2003:
Source: Research Affiliates
T-stats of less than 2 generally indicate that the finding is not statistically significant. The fact that none of the factors have t-stats above 2 for the 2003-2018 period show that the any excess returns are statistically indistinguishable from zero.
Graphically, we can see that a portfolio of the first 6 and a portfolio of the last 8 factors in the table above have both underperformed beta since they were all well known in 2003:
Source: Research Affiliates
Research by Jacobs and Muller confirmed the findings of the post-publication decline in factor returns in the US. However, when they turned their sights on international markets outside the US, they found that “none of the 38 international markets in our sample yields a reliable post-publication decline in anomaly returns”.
Looking at the table of their findings below, the t-stats in brackets show that the findings are all significant. Factor returns in the US show in-sample returns of 0.74%, which reduce to 0.47% post-sample, which reduce to 0.29% post-publication. This is consistent with what we’ve already seen. But looking at international markets, in contrast the authors find no deterioration in returns either post-sample or post-publication (0.41% in-sample, 0.50% post-sample, 0.52% post-publication).
Source: Jacobs and Muller
The authors argue that factor deterioration might occur in the US, but not elsewhere, because of the presence of a number of barriers within international markets preventing them from exploiting factors. They suggest that such barriers may include short-selling being more expensive in international markets, political risks, and/or increased regulatory restrictions.
Given that I live in the UK, I have a natural interest in how things apply to the UK. Whilst there are almost no UK factor funds to invest in, I’m still curious about how factor returns hold up. Although the Jacobs and Muller paper finds nothing statistically significant from their testing (they break down the returns by country in Table 3), the next paper looks specifically at post-publication declines in factor returns for the UK.
Research from University College Dublin found a similar phenomenon to the original studies conducted on US data. Sadly for those people who have managed to invest in UK factors, the authors find that their results on factors in the UK “show a general decline in the significance of well-known anomalies in the UK stock market. This is consistent with an improvement in market efficiency over time with respect to well-known anomaly variables.”
Are factor returns going to disappear?
Given the declining efficacy of factor returns, one could ask whether factors run the risk of being eliminated completely. However, thanks to the nature of factors, it’s unlikely that the major factors’ excess returns will totally disappear. And the reason is one we’ve already seen. In my article on why factors exist, we saw that their existence was likely a combination of 1) some stocks being riskier than the market, combined with 2) investor behaviour causing pricing inefficiencies. The benefit of factors having a risk-based element to them means that they’ll always be riskier, regardless of how much capital is allocated to them. Risky stocks remain risky even if everyone knows about them.
Take the equity risk premium (ERP) for example – or the ‘beta’ factor as we could also call it. This is a good example of a factor existing solely because of risk (and not behaviour). We all know that over the long run, we’ll get excess returns by allocating to the market over the risk-free asset, regardless of how many other people do the same. Nobody believes the ERP will disappear because everyone knows about it.
However, most other factors also contain a behavioural element, which is susceptible to being competed away as those making the behavioural mistakes wise up and stop making them, and more people start to exploit the behavioural mispricings. It therefore seems prudent to assume that at least part of a factor’s excess return discovered by academics will be competed away by other market participants. AQR also take this view, noting that factors unlikely to be completely arbitraged away, but that it’s a good idea to assume lower returns to factors following their widespread adoption.
Given that most investors will be getting their factor exposure from US equities (as there are drastically fewer ex-US factor funds), the deterioration of factor returns in the US proves troublesome for most investors.
Overall, I think it makes sense for investors to go into factor investing without their glasses being rose-tinted by the high historic returns advertised in academic research. The evidence points to factors continuing to exist, but providing lower returns going forward thanks to the widespread adoption of factor-based strategies. Factors may still provide some benefit to portfolios, but investors should manage their own return expectations when allocating to these sorts of strategies.
The next post in the series focuses on how difficult it can be to tell when a factor stops working.