So, I was in Boston talking about low volatility to a couple of groups. I met a couple of current practitioners and it is very interesting to hear about the different approaches these guys take. They tend to have very different takes on what is going on, and what investors really care about, and these subtle differences lead to very different portfolios if you are talking about choosing 100 stocks out of 2000 possible.
More cynically, each firm selling these approaches needs to differentiate itself, because simply trying to be better than a competitor is risky: there's at least a 50% chance of failure, and possibly higher because it might degrade into a simple price war that leads to no prices (Bertrand competition). There's no need to be so cynical here because everyone comes to low vol from such a different approach (eg, flat or negative Security Market Line?).
At my Qwafafew talk (during which, I quaffed a few--beer and finance are complements), there was a true believer in the audience defending the conventional wisdom, which was actually quite refreshing. I like actually engaging with them because they generally don't argue with me, adopting the debating tactic of hoping an embarrassing criticism will go away if you ignore it. They have been doing that to me for years. Anyway, he insisted that beta was positively correlated with average returns, you merely had to measure beta correctly. I can't prove it can't be done, but I haven't seen it, and I'm pretty certain any journal would publish such a result asap, so I'm skeptical.
I'm even more skeptical because when I asked him if he would agree that beta is positively correlated with total variance, he said, 'not necessarily!' as if this was some great gotcha point. Correlations refer to statistical relationships, meaning, there are some datapoints off the linear relationship. Clearly, some stocks will have high variance and low beta or vice versa, but all discussions about risk and return are about averages and tendencies; the distinction about there being exceptions to financial generalizations is not like saying there are exceptions to gravity. I thought he was really reaching if that's the kind of counterargument to my assertion that risk and return aren't correlated in general.
He did bring up the point that I tended to put up data on geometric averages, and that arithmetic averages would generally be more consistent with the theory because the arithmetic returns are greater than their geometric averages by a variance term. That's a valid point, but only geometric averages uniquely correspond to total period returns, and so the geometric average is to my mind a better statistic of a portfolio's performance. Further, this is why the scope of my findings are so important: across many asset classes, the average returns are generally flat. Average returns are unbiased estimates of future total returns, but then sometimes one should see massive outperformance by these high volatility portfolios, which we do not. It's a complicated issue, and I can't treat it sufficiently in this point, but I should mention that while I think this is one of the best arguments by the risk-begets-return crowd, it still fails empirically.
Then there's the issue of overcrowding that always seems to come up. I find this rather amusing, because when I was trying to sell this idea in the 1990's, before anyone thought it would work, a common argument was that 'what if everyone did it?' It's like when you show someone something that has worked well for 50 years, and they say, 'sure, but now it can't work because I'm sure everyone sees it.' Well, if everyone invests by maximizing Sharpe ratios, then we are in the CAPM world that creates the positive sloping security market line, but we've been telling MBAs to do this for decades to no avail. Considering all the money in value and small cap funds, I don't think the modest allocation to low volatility has changed things.
In fact, there are two good pieces of research put out on this point recently. One by Pim van Vliet at Robeco (download here) documents a metric of value for low vol portfolios back to 1930. He finds low vol has gotten more expensive over the past 5 years, but looking at it in context it is not so scary (it has been here before). Nevertheless, he argues this is why you need his secret sauce. Self serving, but rightfully so. Another (here), by quants at Deutsche Bank, found that low vol is not overcrowded primarily using pairwise correlation data. That's an interesting way to view crowding.
One thing I found interesting is that in institutional asset management, finance and econ PhDs are common. That's quite different than in my day-to-day field where physics and computer science PhDs are more common, where we aren't interested in soliciting client money. I'm on the fence as to whether this means financial theory is useful for investing large amounts of assets in broad asset classes...or that it merely is more helpful in selling one's services to institutions.
More cynically, each firm selling these approaches needs to differentiate itself, because simply trying to be better than a competitor is risky: there's at least a 50% chance of failure, and possibly higher because it might degrade into a simple price war that leads to no prices (Bertrand competition). There's no need to be so cynical here because everyone comes to low vol from such a different approach (eg, flat or negative Security Market Line?).
At my Qwafafew talk (during which, I quaffed a few--beer and finance are complements), there was a true believer in the audience defending the conventional wisdom, which was actually quite refreshing. I like actually engaging with them because they generally don't argue with me, adopting the debating tactic of hoping an embarrassing criticism will go away if you ignore it. They have been doing that to me for years. Anyway, he insisted that beta was positively correlated with average returns, you merely had to measure beta correctly. I can't prove it can't be done, but I haven't seen it, and I'm pretty certain any journal would publish such a result asap, so I'm skeptical.
I'm even more skeptical because when I asked him if he would agree that beta is positively correlated with total variance, he said, 'not necessarily!' as if this was some great gotcha point. Correlations refer to statistical relationships, meaning, there are some datapoints off the linear relationship. Clearly, some stocks will have high variance and low beta or vice versa, but all discussions about risk and return are about averages and tendencies; the distinction about there being exceptions to financial generalizations is not like saying there are exceptions to gravity. I thought he was really reaching if that's the kind of counterargument to my assertion that risk and return aren't correlated in general.
He did bring up the point that I tended to put up data on geometric averages, and that arithmetic averages would generally be more consistent with the theory because the arithmetic returns are greater than their geometric averages by a variance term. That's a valid point, but only geometric averages uniquely correspond to total period returns, and so the geometric average is to my mind a better statistic of a portfolio's performance. Further, this is why the scope of my findings are so important: across many asset classes, the average returns are generally flat. Average returns are unbiased estimates of future total returns, but then sometimes one should see massive outperformance by these high volatility portfolios, which we do not. It's a complicated issue, and I can't treat it sufficiently in this point, but I should mention that while I think this is one of the best arguments by the risk-begets-return crowd, it still fails empirically.
Then there's the issue of overcrowding that always seems to come up. I find this rather amusing, because when I was trying to sell this idea in the 1990's, before anyone thought it would work, a common argument was that 'what if everyone did it?' It's like when you show someone something that has worked well for 50 years, and they say, 'sure, but now it can't work because I'm sure everyone sees it.' Well, if everyone invests by maximizing Sharpe ratios, then we are in the CAPM world that creates the positive sloping security market line, but we've been telling MBAs to do this for decades to no avail. Considering all the money in value and small cap funds, I don't think the modest allocation to low volatility has changed things.
In fact, there are two good pieces of research put out on this point recently. One by Pim van Vliet at Robeco (download here) documents a metric of value for low vol portfolios back to 1930. He finds low vol has gotten more expensive over the past 5 years, but looking at it in context it is not so scary (it has been here before). Nevertheless, he argues this is why you need his secret sauce. Self serving, but rightfully so. Another (here), by quants at Deutsche Bank, found that low vol is not overcrowded primarily using pairwise correlation data. That's an interesting way to view crowding.
One thing I found interesting is that in institutional asset management, finance and econ PhDs are common. That's quite different than in my day-to-day field where physics and computer science PhDs are more common, where we aren't interested in soliciting client money. I'm on the fence as to whether this means financial theory is useful for investing large amounts of assets in broad asset classes...or that it merely is more helpful in selling one's services to institutions.
3 comments:
Actually, you'd be surprised at the difference between beta and volatility.
Small-cap biotechs are low-beta, high-vol. Many energy stocks are high-beta, low-vol.
Run a screen on the two. It's interesting to see where the differences lie.
And I'd argue that low-beta, high-vol, small-cap biotechs are likely to underperform because of the lottery ticket bias. The same cannot be said of high-beta, low-vol energy stocks.
"I think it reflects the fact that for longer term investing."
Is there something missing from this sentence?
One story that could be told for prospective continued outperformance of low(er) vol portfolios would be that as the US proportion of world GDP shrinks over time, the tastes of non-US investors will become increasingly influential in the US shares market, and those investors, e.g. from BRIC countries, are likely to prefer familiar, high-cap brand-name dividend payers, much as they focus their real-estate buying in the internationally-known metropolises such as NYC rather than in outlying areas.
Post a Comment