It has been a while since I wanted to write about the “low vol” factor. The reason is pretty simple, look how it performed in the last year (white line is the biggest LowVol ETF listed in the US, the blue line is SPY):
I invest in LOWV, the UCITS version managed by State Street, and thanks to the currency exposure, its graph is even more impressive:
I did not show you the one-year performance because I have a very special course/pdf to sell and I decided to go all-in on recency bias; I did it because I am stupefied no one is doing it. Normally in finance, recent winners are glorified way above justification while, sitting in front of my little window on the FinWorld, I do not see any hype. Is my window pointing in the wrong direction?
The Low Vol factor
The are a lot of articles out there that explain better than I could what this factor is and where it comes from; I do not want to write the Nth version but since I know no one clicks on links, here is a short version of it. According to the CAPM, higher risk should be compensated with higher returns but empirical studies show that stocks with a low beta, a measure of systematic risk exposure, overperform stocks with high beta. Why so? Most likely for the following two reasons:
- investors have leverage constraints and leverage aversion
- skewness preference: many investors want a lottery-like payoff.
This is why “no one eats risk-adjusted returns” is such a good maxim. Investors want high, not risk-adjusted high, returns because that’s what ultimately is useful for their goals. And their preferred way to get them is via stocks that, at least on paper, promise to deliver those high returns. Access to cheap leverage is constrained but, even if it was available for everyone, not every investor would use it. Just think about all those obsessed with repaying their mortgage as fast as they can.
TLDR: the average investor prefers high-octane stocks over levered high risk-adjusted returns (or they would like the latter but cannot access cheap leverage).
From asset class diversification to factor diversification
In building financial portfolios, asset class diversification is the rule of the game. Does it represent the best possible approach to the problem though?
I mean, the debate about geographical diversification is not close yet…as if a company’s decision to list on a stock exchange in Madrid or Seoul should drive its P&L or investors have a superior ability to forecast returns of their local stock market.
I recently stumbled on a paper from Antti Ilmanen and Jared Kizer that offer some (stale? the paper is dated 2012) food for thought. According to the authors, there is ample room for improvement by shifting the focus from asset class diversification to factor diversification. The improvement in risk-adjusted returns mainly comes when investors need it most, during crisis periods, because it is in those instances that asset class correlations rise, as the market switches between binary risk-on / risk-off environments. With traditional portfolios, diversification fails in short-term panics (and by short term here they mean some years) but effectively reduces downside risk over longer horizons; factor diversification is simply a more effective way to introduce risk diversification in financial portfolios.
Any asset can be viewed as a bundle of factors that reflect deeper risks and rewards; factors are not just a prerogative of stocks. For this reason, the paper does not cover only value and momentum in stocks but also carries in fixed-income and currency markets and trend-following strategies in all asset classes. The authors also shift focus from dollar allocation to risk allocation (same principle as risk parity). All these considerations bring the correlation between factors even lower.
The paper shows that the average correlation between the five, equally weighted, basic constituents of the Global Asset Allocation portfolio (US and DM stocks, US and DM Govies, and a final ensemble of small caps, EM stocks, property and commodity) is .38 while the average correlation between four factors (value, momentum, carry and trend) and large US stocks is virtually 0 (-0.02).
Here is the most important conclusion of the paper: “Better tail performance in equity bear markets: during the 46 worse months for US stocks (10% of the sample), the asset-class diversified portfolio lost 3.8% on average while the factor-diversified portfolio lost only 0.9%. During the other 90% of the sample, both portfolios earned a 1.3% average monthly return.“
For someone like me that use factors in his stock portfolio, this sounds great. Now the bad news…
Unfortunately all the above is valid for long-short factors, meaning for example that the value factor is derived going long stocks with low valuations and short stocks with high valuations. There is no ETF out there that offers similar exposure to the value factor (for sure not in Europe). If we look at the long-only value and momentum factors and the related ETFs we can invest in, they lose plenty compared to their long-short equivalents: their once negative correlation (-0.53) becomes 0.73 and both factors also result heavily correlated with the US stock market.
In the long-only case, the better risk-adjusted performance of the factor portfolio over the asset class portfolio is largely (two-thirds) explained by the better performance of trend and carry relative to bonds and alternatives as complements to the all-equity portfolio. The improvement of using value and momentum instead of the standard market-cap-weighted equity portfolio becomes quite small.
The Real World
I have taken three ETFs to replicate factors, MTUM for momentum, IVE for value and SPLV for low volatility, and went to work. Here is their correlation to the S&P500, represented by the SPY ETF, according to PortfolioVisualizer:
The first issue we have in the ‘real world’ is that factor ETFs are relatively young. There is not much history and, more crucially, a variety of market environments to work on. At least the data we have seem to confirm the paper’s conclusion: long-only factors are highly correlated with the S&P500. Also to note, the factor with the lowest correlation is LowV.
What I am curious to test is if using long-only factors instead of the standard SPY I can improve the risk-adjusted returns of some of the “classic portfolios” out there. Given that I am not exactly drowning in free time at the moment, I started with the OG, the 60/40.
Unfortunately, PortfolioCharts.com supports only the value factor in its portfolio builder feature and, despite several attempts over the years, I never managed to nail how to use the PORT function in Bloomberg. So…here we are again, PortfolioVisualizer.
According to the optimizer, this is the 60/40 configuration (with the AGG allocation locked at 40%) that generated the lowest drawdown since 2014:
Not surprisingly, it allocates to the two factors that have the lowest correlation: SPLV and MTUM.
Here is the standard 60/40 over the same period for reference:
The factor portfolio improves the Sharpe ratio from 0.76 to 0.82, a resounding “meh“. But at least seems to point in the right direction; the good news about the current bear market is that every passing day it adds valuable data for future stress testing analysis.
I also did a couple of experiments with Composer. The aquamarine (?!?) line represents an equal-balance portfolio of SPLV, MTUM and IVE. The purple line is a portfolio that invests 100% in the factor that had the best cumulative return in the 200 days prior, a sort of momentum of factors.
My actual experience with factors taught me that they experience long periods of over or underperformance. Assuming that the three of them would all individually produce higher risk-adjusted returns compared to the S&P500, I thought that the factor-weighted portfolio would take advantage of a sort of mean-reversion effect, buying the low-performing factor while selling the high. The numbers seem to disagree, the performance is basically indistinguishable from the main index.
This is not the only, nor the best, way to build a multi-factor portfolio. Taking this road I really risk ultimately owning the market…but with higher fees and turnover. As a reference, pros didn’t nail it either:
The factor momentum thought…turned out to be just a back-test fluke. If you change the rebalancing frequency or the lookback period, the results are very similar to SPY.
Again, maybe factors are not proving their worth because of the short back-testing period we can work with. According to the paper, their outperformance shine during the worst stretches for stocks and since 2014 we had the Covid crash and now, that’s it.
AQR in the Real World
Out of curiosity, I went to check some AQR funds (Ilmanen works there) to see if making your own sausage factors led to the desired results.
The AQR Style Premia Alternative Fund employs a market-neutral, long/short strategy across four investment styles (value, momentum, carry and defensive) and five asset groups (stocks, fixed income, currencies, equity indexes and commodities).
As you can see from the above graph, for sure the fund is not correlated with the S&P500. But that’s where the good news stops. It is difficult to compare its returns to the one in the paper, given that the methodology and ingredients are not the same and the track record is not that long but…sub 4% p.a. is not exactly what I was expecting (with that volatility, on top).
As a standalone fund, or as 80% allocation next to 20% SPY (as suggested in the paper), ain’t no great investment.
What if we put it with the 60/40, like QSPIX 25, SPY 45, AGG 30? That’s the Portfolio1 in below image while Portfolio2 is the standard 60/40:
The result is not bad at all but…aren’t we back at square one? Weren’t we supposed to abandon the asset allocation logic?
Bottom Line
There is a great passage in the book Becoming Trader Joe: “It’s better to have a good plan and stick with it than having a great plan and continue to tinker around it”.
Using factors instead of the plain vanilla SPY looks like the tinkering plan. Or maybe we didn’t have yet a market disaster big enough to appreciate what factors can bring to a portfolio.
To not leave you thinking “I wasted ten minutes of my life reading a random idiot dunking on AQR”, I give you this gem that popped on my Twitter feed:
Imagine being up 6% in 2022 using two funds and one is a levered stock&bond ETF 😉
The Humpty Dumpty Portfolio
def: an unnecessarily complicated portfolio, usually adopted by financial advisors to justify above-market fees.
It is true that all factors combined might give you back the same return as the S&P500. But if you get that when times are good, you also get a free option to possibly outperform the benchmark when shit hits the fan. Paying more to complicate your portfolio can be a waste but if you get that allocation for free, why not?
The added issue here is that I want to get exposure to the value factor but I do not know which ETF provider has THE exposure to it. Should it stick to the Fama-French definition? Should it include intangibles? What about the optimal lookback period for momentum? And so on…
Someone can look at a portfolio with two value ETFs and conclude humpty-dumpty, someone else can see it as a form of model-risk diversification. Am I able to tell you if the Alpha Architect lads found a better formula than Meb Faber? How long should be the period to evaluate them? What if any of them change their model mid-way? Few of us have the luxury of being able to build and test our own model like AQR.
Now apply the same logic to something more valuable, from a portfolio return POV, but more complex like trend following and boom…diversify your diversifiers into N=infinite allocations!
Good luck
What I am reading now:
Follow me on Twitter @nprotasoni