More and more investors are seeking exposure to a number of particular risk factors in their equity portfolios, driven by extensive academic evidence that such factor investing exposures will be rewarded with premiums.
To implement factor investing strategies, they typically invest in equity index funds or ETFs that are tilted towards the cheapest (value), most profitable and better managed (quality) companies with the lowest risk and the strongest price and earnings trends (momentum).
How you allocate to these stocks can make all the difference. As we have shown in our most recent paper, carefully removing unwanted exposures to other non-rewarded risk factors increases the risk-adjusted excess returns of factor investing strategies. We call this the purification of factor premiums.
What also helps is the diversification of information used to assess how cheap, profitable, well-managed or low risk a company is, or how strongly its market price and earnings are trending. We call this the diversification of factor premiums.
Finally, dynamically adjusting the active share of the factor investing portfolio so that the fund’s overall volatility of the excess returns relative to the benchmark index, i.e. tracking error, is kept constant over time, also leads to higher risk-adjusted returns, something we had already discussed in a previous paper.
Our research results are available in chapter 4 of the recent book on “Factor Investing, From Traditional to Alternative Risk Premia” edited by Emmanuel Jurczenko, Associate Dean and Professor of Finance at Ecole Hôtelière de Lausanne, and published last month by ISTE Press – Elsevier.
The problem with over-simplistic approaches
Academics tend to use relatively simplistic approaches to demonstrate evidence of factor premiums. For example, Professor Ken French maintains a data library on his website showing the performance of simple value (HML) and momentum (Mom) equity strategies. This is often used by academics and practitioners to assess fund exposures to the value and momentum factors, i.e. how much of the returns to a fund can be explained by tilts to value or momentum stocks.
The HML value factor simply prefers the cheapest stocks, with the highest book-to-market, to the most expensive stocks, those with the lowest book-to-market. The Mom momentum factor prefers the stocks with the strongest returns over the last 12 months (excluding the most recent) over those with the weakest momentum based on the same measure.
Such preferences would translate into overweights and underweights of stocks in a fully invested portfolio. And the returns provided by Ken French on his website can be seen as the excess returns that would have been generated by those stock overweights and underweights.
The HML value and the Mom momentum factors use only one source of information, i.e. there is no diversification of the information used to assess how cheap or trending a stock is. Also, there is nothing in Ken French’s methodology to ensure sector neutrality or to explicitly neutralise an exposure to the market portfolio (beta).
The result is that the returns to the HML value factor and Mom momentum factor available in Ken French’s website can be driven by other risk factors such as sector-active exposures, or correlated with the returns to the market portfolio. Finally, Ken French’s methodology uses nothing to dynamically change the active share so that a constant level of active risk is targeted over time. The volatility of the returns to the HML value factor and Mom momentum factors can thus vary greatly over time.
Unfortunately, simplicity comes at a cost, as shown below, and HML and Mom are poor representatives of what asset managers should apply to their equity portfolios to create exposures to value and momentum factors.
Uncontrolled market exposure in factors
Exhibit 1 shows the market exposure measured by the 3-year rolling active beta (the beta of the fully invested portfolio subtracted by one) of the portfolio strategies behind the HML value and Mom momentum factor returns for US stocks, using data from Ken French’s website.
A reading of 0 corresponds to a neutral active beta, i.e. no market exposure generated by the active stocks tilts, and thus no correlation of the market returns with the portfolio returns in excess of the benchmark index is expected.
For Mom, the active beta would have been as high as 1 in 1943 and 1966 and as low as -0.95 in 1940, changing wildly over the years.
For HML, the active beta changed less over time but still reached 0.92 in 1944 and -0.65 in 2001, staying above 0 for most of the 30s and 40s and below 0 from the 60s and until recently.
This means that both HML and Mom excess returns over benchmark can be partially explained by their frequently high positive or negative correlation with market portfolio returns.
Exhibit 1: Active beta of the HML and Mom factors
Source: BNP Paribas Asset Management and data library on Ken French’s website
That has huge implications. Take Mom for example. Between the 22 September 2011 and 21 September 2012 the average active beta is -0.49. The active beta contribution to Mom is thus -14.50% if we take into account the positive average return of US equities (in excess of interest rates) in that same period. Since the average excess returns generated by the Mom portfolios in the same period was only +0.08%, that means that if the market exposure had been hedged (setting the active beta to zero) investors would have earned +14.58% in excess returns instead, i.e. the actual alpha in the Mom factor.
That means the Mom factor paid a positive premium in that same period but this was masked by the fact that the underlying portfolio was not purified to remove an unwanted negative correlation with the returns to the market portfolio. In fact, a positive uncorrelated component of the Mom premium was completely washed away by the impact of that negative correlation with market returns and the fact that the market portfolio moved adversely.
This illustrates well that removing unwanted risk exposures from strategies designed to capture factor premiums should have a positive impact on the risk-adjusted performance in the long term. Other unwanted risk exposures include market capitalisation biases, sector biases and, in global portfolios, regional biases.
Uncontrolled volatility of factor excess returns
Exibit 2 shows the 3-year rolling historical volatility of HML value factor and Mom momentum factor excess returns for US stocks using data from Ken French’s website. This can vary enormously for these two factors, from 3.3% in 1977 to over 27% in 2008. That simply means that changes in HML and Mom monthly excess returns from one month to the next were small in the 70s but significantly larger 2008 and 2009. Thus, the 70s will have a much smaller weight in the average excess returns to the HML value factor and the Mom momentum factor than years like 2008 and 2009.
Exhibit 2. Ex-post volatility of the HML and Mom Factor excess returns
Source: BNP Paribas Asset Management and data library on Ken French’s website
As we demonstrated in our earlier paper, the volatility of the HML value and the Mom momentum factor excess returns can be predicted with great accuracy. Thus, by changing the active share in benchmarked portfolios, i.e. changing the extent of the tilt towards value and momentum stocks, one can target a constant tracking error over time. In that same paper we also demonstrated that the factor premiums of HML, and in particular those of Mom, are negatively correlated to their respective tracking error. For that reason, managing the active share so that exposures to the HML value and Mom momentum factor are reduced when their respective tracking error is increasing and the average excess returns are decreasing, and conversely, results in higher risk-adjusted excess returns, a simple timing effect as explained in more detail in our previous blog.
This demonstrates the importance of managing the active share in benchmarked portfolios so as to target a constant tracking error over time.
Diversifying sources of information in factor investing
A great amount of academic literature discusses value, quality, low risk and momentum. However, there is no consensus about what indicators should be used in order to decide whether a company is cheap, profitable, well managed, or less risky, or the extent to which a company’s price or its earnings is trending. For that reason, in our research we used a variety of the indicators proposed in academic literature (see Exhibit 3 below).
Because of the diversification effect, risk-adjusted excess returns are higher when more information is used, in particular when the most diversifying sources of information behind each factor are included. In the paper we discussed the increase in risk-adjusted excess returns arising from this effect, too.
Exhibit 3: The list of all factors used in our analysis
By how much can the risk-adjusted factor premiums increase?
The benefits of removing unwanted risk exposures from factor investing strategies, targeting constant tracking error and diversifying information sources are huge. In Exhibit 4 we first show the expected average information ratio for value, quality, momentum and low risk investing when using only one indicator, keeping the active share constant over time, and refraining from hedging the market beta, the sector exposures, the regional exposures and the size exposures. We call this ‘over-simplified’.
We then show the information ratios for robust factor strategies that equally weight all sources of information in Exhibit 3 and change the active share dynamically so that the tracking error is kept constant. We also neutralise the market beta and the sector and regional exposures, and make sure there are no significant market capitalisation biases in the portfolio. We call this ‘purified and diversified’. The increase in information ratio is significant in all cases. The last row shows the impact on the multi-factor portfolio including all value, quality, momentum and low risk sources of information in Exhibit 3 in one single strategy. The results are remarkable.
Exhibit 4: Simulated information ratios for over-simplified and for robust factor strategies demonstrating the huge importance of removing unwanted risk exposures, managing total volatility and diversifying sources of information.
NB: World: stock universe defined by the MSCI. World index from January 1997 to November 2016, results in USD. US: stock universe defined by the S&P 500 index from January 1990 to November 2016, results in USD. Europe: stock universe defined by the STOXX Europe 600 index from January 1992 to December 1998; results 1990-1998 in German D-Mark and from January 1999 to November 2016 in EUR. Japan: stock universe defined by the Topix 500 index from August 1993 to November 2016, results in JPY.
Source: BNP Paribas AM, FactSet, MSCI, S&P, Topix, Stoxx, Worldscope and Exshare
Factor investing: the devil is in the detail
The philosophy behind factor investing is not complicated. One seeks to tilt portfolios towards the cheapest, the most profitable and well-managed companies with the lowest risk and strongest price and earnings trends. Not only does this seem intuitive from an investment philosophy point of view, it has also been shown by academics to be a way of targeting higher returns from equity markets.
But the portfolio construction and diversification required to allow investors to profit from these tilts is not simple. Investors not paying attention to how carefully factors are diversified and purified are in for a disappointment when engaging in factor investing.