Okay, so we shifted a comma over a little. But this week’s Economist contains a column that caught our eye here at All About Alpha. The piece, entitled “What’s it all about, alpha?” acknowledges the rising importance of alpha-centric investing:
“It is the fashion these days to separate beta (the systematic return delivered by the market) from alpha (the manager’s skill). Investors are happy to pay high fees for the skill, but regard the market return as a commodity. Distinguishing the two is, however, difficult.”
The column goes on to question whether too much beta is even a problem since there is naturally a lot of value in he selection of the right betas.
“…alpha sceptics often attribute eye-catching returns to style bias, such as favouring stocks with a high dividend yield.
“But should they be biased against style bias? After all, the only portfolio utterly free of bias would be one that included the entire market.”
Not quite true, we contend. The only portfolio that is utterly free of bias is the one the investor chose themselves. In other words, if an investor – for whatever reason – chooses to invest in technology companies, then a technology fund would be “free of bias”. Ironically, a fund that included the entire market would represent an active bet against technology and for non-technology sectors. As we have argued before, beta is in the eyes of the beholder. Put another way, one person’s beta is another person’s alpha.
The central problem with factor models is that they are built using a best fit analysis on historical data. They are designed specifically to reduce the amount of unexplained returns in historical data. Try this experiment in Excel: Create a column of 20 random numbers between one and 100 [“=RAND()*100”]. Copy and “past>special>values” elsewhere in your worksheet. When you do, the first set of numbers will change. Calculate the correlations between the cut & pasted column and original one. If you’re really into Excel, then go ahead and do a scatter plot and have Excel place the regression equation right on the chart.
You will likely see a non-zero correlation between your (totally independent) columns of data – and non-zero beta. Now imagine that one column is a hedge fund and the other is an equity index – say, the S&P500. Even though the “r-squared” on this relationship might be low, you might still erroneously conclude that part of the hedge fund might have been replicated by simply buying an S&P500 ETF.
Looking forward, things might change though. A relationship that might have existed in the past (either by random fluke as in this example, or due to a certain set of circumstances unique to the last 20 months) might not extend into the future. In this unique scenario, we know for a fact that there is no relationship between the fund and the S&P500. In other words, all returns must essentially be alpha – even if the manager happened to track the index and the fund showed some beta.
This is a dramatic oversimplification of how factor modeling works. Obviously academics test for this problem by running their models on sub-sets of data to see if they hold up. But it tells the same tale: things change. Just because a return stream has a certain correlation with markets in the past does not necessarily mean it will have the same correlation going forward.
For example, riding energy stocks over the past 3 years could be either a) cheap, run-of-the-mill energy beta or b) a shrewd call on energy markets. It’s impossible to really know. Of course, if energy continues to rise year after year for 20 years and a manager tries loading up on energy and calling it alpha, you might know better. But if that happened, the market would have clued in that energy was always going to rise continuously and would have bid those stocks up to (discounted) 2027 levels and no returns would have been made.
Uncomfortably for the hedge fund community, this leads back to Prof. Harry Kat’s hypothesis that hedge fund returns are simply the specific returns demanded by the market for assuming hedge funds’ particular return distributions (measured by volatility, skewedness, kurtosis and correlation). In other words, hedge funds are “fair value” for what you get, not “a free lunch” after all.
At the end of the day, if all alpha eventually evolves into beta, then alpha is transient – what a Ph.D. friend of Alpha Male quite adroitly calls an “ingenuity gap”. The Economist seems to essentially agree, suggesting that clones may never really be able to cross this gap:
“…the clones will always be a step behind the smart money. You cannot clone a hedge fund until you know where it has been.”