The negative alpha of the Value Factor
The negative alpha of the Value Factor
William Bernstein in this post (http://www.efficientfrontier.com/ef/0adhoc/fi.htm) made it quite obvious that value investing comes with negative alpha, and growth investing comes with positive alpha.
The summary of the article in my mind is:
From 19262016 the value factor premium has been 4.8%. However most value funds do not have a factor value of 1. For instance from 19622004 the Fama French Large Value Index has a value factor of 0.75 reducing its outperformance to a theoretical 3.7%. The Fama French Large Value Index also has a negative alpha of 1.5%/year causing actual outperformance to be 2.2%. The Fama French Large Growth Index has a positive alpha of 1.7%/year, so the difference in actual performance between Large Value and Growth is 0.5%/year.
I went to portfolio visualizer and looked up the Vanguard Small Cap Value Index (VISVX). It supports the above thesis by showing a negative 1.7% yearly alpha from 19982017. But most funds I throw at (except large growth) also have negative alphas so I am not sure if its value or small phenomenon.
Even if the there were no negative/positive alphas, I also ponder the question of the significance of the 4.8% annual return premium. From 19291978 (a worst case scenario) the CAGR of S&P 500 was 7.7%. From 19501999 the CAGR was 13.6%. For a 50 year period, this is a 5.9% difference for the same asset class depending on starting date. The yearly standard deviation of S&P500 stocks over the last century is about 18%. A 50 year standard deviation is 18*(1/sqrt(50))=2.5%. Two standard deviations results in a 5% variance. So if the true return was 10%, the variation of returns over 50 years could be 515%. Therefore the value premium of 4.8% may not be statistically significant (Interestingly, if you go back to 1982, and cover the 19321982 time period, the value premium seems robust).
I currently own about 1/3 of my stocks in vanguard value indexes, and they have done fine this year (though not as well as the S&P 500). I do not intend to change course right now. However, I would appreciate knowing more about this negative value alpha as well as any information you know about the statistical significance of the value factor.
The summary of the article in my mind is:
From 19262016 the value factor premium has been 4.8%. However most value funds do not have a factor value of 1. For instance from 19622004 the Fama French Large Value Index has a value factor of 0.75 reducing its outperformance to a theoretical 3.7%. The Fama French Large Value Index also has a negative alpha of 1.5%/year causing actual outperformance to be 2.2%. The Fama French Large Growth Index has a positive alpha of 1.7%/year, so the difference in actual performance between Large Value and Growth is 0.5%/year.
I went to portfolio visualizer and looked up the Vanguard Small Cap Value Index (VISVX). It supports the above thesis by showing a negative 1.7% yearly alpha from 19982017. But most funds I throw at (except large growth) also have negative alphas so I am not sure if its value or small phenomenon.
Even if the there were no negative/positive alphas, I also ponder the question of the significance of the 4.8% annual return premium. From 19291978 (a worst case scenario) the CAGR of S&P 500 was 7.7%. From 19501999 the CAGR was 13.6%. For a 50 year period, this is a 5.9% difference for the same asset class depending on starting date. The yearly standard deviation of S&P500 stocks over the last century is about 18%. A 50 year standard deviation is 18*(1/sqrt(50))=2.5%. Two standard deviations results in a 5% variance. So if the true return was 10%, the variation of returns over 50 years could be 515%. Therefore the value premium of 4.8% may not be statistically significant (Interestingly, if you go back to 1982, and cover the 19321982 time period, the value premium seems robust).
I currently own about 1/3 of my stocks in vanguard value indexes, and they have done fine this year (though not as well as the S&P 500). I do not intend to change course right now. However, I would appreciate knowing more about this negative value alpha as well as any information you know about the statistical significance of the value factor.
Re: The negative alpha of the Value Factor
Wall Street creates, maintains and builds upon an entire language, shorthand if you will, for describing investing strategies. Most disciplines do this so they can communicate with each other in an efficient manner. Some attempt to use it to make the outsider feel like they've got less knowledge in respect to the discipline and that the insiders are experts that can assist. Wall Street is like this and many believe its bull$%#t.
In simple terms, a positive alpha of 1.0 means the fund has outperformed its benchmark index (Total Market Index, for example) by 1 percent. In order to achieve this, since the marketplace is a zerosum game, this special fund will have to take its gains from other investors that hold a fund with a strategy contrary to theirs and who suffer corresponding "negative alpha". (They cannot take gains from the Total Market Index because those investors have achieved exactly what they were promised and in this case they are simply the benchmark.)
The problem with Value Funds (as Benjamin Graham learned long ago) or with any other alpaseeking approach is that the contrary fund investors learn after licking their wounds and respond with a better strategy. In Wall Street parlance, this is called efficiency.
Any strategy designed to beatthemarket might do so for some period of time, but very rarely for a long time. Over the long term a market index is better. At least this is what the Bogleheads believe and back up in their own way with historical data. The analysis here is described by Jack Bogle and others.
In simple terms, a positive alpha of 1.0 means the fund has outperformed its benchmark index (Total Market Index, for example) by 1 percent. In order to achieve this, since the marketplace is a zerosum game, this special fund will have to take its gains from other investors that hold a fund with a strategy contrary to theirs and who suffer corresponding "negative alpha". (They cannot take gains from the Total Market Index because those investors have achieved exactly what they were promised and in this case they are simply the benchmark.)
The problem with Value Funds (as Benjamin Graham learned long ago) or with any other alpaseeking approach is that the contrary fund investors learn after licking their wounds and respond with a better strategy. In Wall Street parlance, this is called efficiency.
Any strategy designed to beatthemarket might do so for some period of time, but very rarely for a long time. Over the long term a market index is better. At least this is what the Bogleheads believe and back up in their own way with historical data. The analysis here is described by Jack Bogle and others.

 Posts: 2390
 Joined: Fri Feb 23, 2007 8:21 pm
Re: The negative alpha of the Value Factor
What was once alpha turns into beta. Once a factor is revealed by academic research and made popular, it’s expected return will decrease but likely not disappear. The risks persist and human behavior is quite persistent. Once a factor becomes known, the appropriate benchmark changes as well. The single factor model CAPM explained about 2/3 of a portfolio’s returns. Current 4 or 5 factor models explain over 90% of a portfolio’s returns. Alpha implies returns above or below what can be accounted for with current known drivers of returns. There’s not much room for it anymore. When one looks to see if a value fund adds alpha, he needs to account for the value factor exposure, then see if there is any excess return beyond that. I’m not sure that fundamental indexing adds anything beyond its value factor tilt.
Dave
Dave

 Posts: 1758
 Joined: Fri Aug 06, 2010 3:42 pm
Re: The negative alpha of the Value Factor
Factor analysis doesn't paint the whole picture for equity returns. It is instructive to do a factor analysis of VPMAX, Vanguard's Primecap Fund since its inception more than 16 years ago. During that timespan a 4 factor analysis by Portfolio Visualizer reveals its monthly Market factor at 0.00, its Value factor at 0.00, its MOM at minus 0.383/mo, and its Size factor at minus 0.271/month. Based on a 4 factor analysis we would expect its factor based monthly returns to be minus 0.654 per month or minus 7.48% per year relative to beta over that time frame. Bear in mind that some factor adherents claim that around 90% of equity returns are determined by this 4 factor model. So based on this, we would expect VGMAX to massively underperform during its lifetime. In point of fact VGMAX has outperformed massively and consistently for 16 years both its peers both capweighted index and factor approaches ever since its inception. It has hugely outperformed every other Vanguard large cap fund both for 10 years and for 16 years. Many believe that DFSVX is the purest and best SCV fund based on its highly impressive performance since its inception in 1992. In point of fact since its inception 25 years ago, DFSVX has underperformed VPMAX, a large cap growth fund, in spite of the fact that factor analysis looks far more promising for DFSVX. VPMAX's impressive success is clearly not due to factors as we define them, but rather to positive alpha and very skillful managers.
It is important to remember that although factor analysis can be helpful for investors, it is a model and, like all models, it can differ meaningfully from reality. The concept of small value makes sense from both a risk and behavioral point of view. The question is not does it exist. I believe that Buffett and Munger would argue that it clearly exists but that the value indexing model may not always capture it effectively after costs. Also, I believe that they would argue that there is a rare commodity called skill which can sometimes create outperformance that the factor analysis model does not adequately explain.
Garland Whizzer
It is important to remember that although factor analysis can be helpful for investors, it is a model and, like all models, it can differ meaningfully from reality. The concept of small value makes sense from both a risk and behavioral point of view. The question is not does it exist. I believe that Buffett and Munger would argue that it clearly exists but that the value indexing model may not always capture it effectively after costs. Also, I believe that they would argue that there is a rare commodity called skill which can sometimes create outperformance that the factor analysis model does not adequately explain.
Garland Whizzer

 Posts: 1758
 Joined: Fri Aug 06, 2010 3:42 pm
Re: The negative alpha of the Value Factor
Factor analysis doesn't paint the whole picture for equity returns. It is instructive to do a factor analysis of VPMAX, Vanguard's Primecap Fund since its inception more than 16 years ago. During that timespan a 4 factor analysis by Portfolio Visualizer reveals its monthly Market factor at 0.00, its Value factor at 0.00, its MOM at minus 0.383/mo, and its Size factor at minus 0.271/month. Based on a 4 factor analysis we would expect its factor based monthly returns to be minus 0.654 per month or minus 7.48% per year relative to beta over that time frame. Bear in mind that some factor adherents claim that around 90% of equity returns are determined by this 4 factor model. So based on this, we would expect VGMAX to massively underperform during its lifetime. In point of fact VGMAX has outperformed massively and consistently for 16 years both its peers both capweighted index and factor approaches ever since its inception. It has hugely outperformed every other Vanguard large cap fund both for 10 years and for 16 years. Many believe that DFSVX is the purest and best SCV fund based on its highly impressive performance since its inception in 1992. In point of fact since its inception 25 years ago, DFSVX has underperformed VPMAX, a large cap growth fund, in spite of the fact that factor analysis looks far more promising for DFSVX. VPMAX's impressive success is clearly not due to factors as we define them, but rather to positive alpha and very skillful managers.
It is important to remember that although factor analysis can be helpful for investors, it is a model and, like all models, it can differ meaningfully from reality. The concept of small value makes sense from both a risk and behavioral point of view. The question is not does it exist. I believe that Buffett and Munger would argue that it clearly exists but that the value indexing model may not always capture it effectively after costs. Also, I believe that they would argue that there is a rare commodity called skill which can sometimes create outperformance that the factor analysis model does not adequately explain.
Garland Whizzer
It is important to remember that although factor analysis can be helpful for investors, it is a model and, like all models, it can differ meaningfully from reality. The concept of small value makes sense from both a risk and behavioral point of view. The question is not does it exist. I believe that Buffett and Munger would argue that it clearly exists but that the value indexing model may not always capture it effectively after costs. Also, I believe that they would argue that there is a rare commodity called skill which can sometimes create outperformance that the factor analysis model does not adequately explain.
Garland Whizzer
Re: The negative alpha of the Value Factor
It turns out I was mistaken above. After many more regression analysis runs, the "alpha" on different funds seems pretty random, except that high expense ratios tend to cause more negative alpha. There are value index funds with positive alpha, and growth index funds with negative alpha. Even balanced market indexes can have large positive or negative alpha (the extended market index for instance has had 0.54% negative alpha on average for the last 30 years by FF 3 factor).
As garlandwhizzer and random walker point out, the FF model is just a model and it can differ from reality. That VPMAX fund has truly spectacular alpha, but I don't expect a model to give me an active manager's return. However I was hoping diversified indexes would have near zero alpha. I was hoping the FF model would be accurate enough to sniff out poor indexes or those with hidden costs, but its just too inaccurate. I currently see a 2% negative alpha on an index fund and it gives me pause; maybe it shouldn't...
And as Ron Scott notes, the only fund that really hasn't had much alpha (positive or negative) has been the total stock market, but this is only true for the FF 3 and 4 factor models. The wonky FF 5 factor model shows the total stock market with a 0.29% alpha for the past 25 years. Perhaps the expense ratio and trading costs play a role here, but I think the model is slightly off. Thanks for your thoughts guys.
As garlandwhizzer and random walker point out, the FF model is just a model and it can differ from reality. That VPMAX fund has truly spectacular alpha, but I don't expect a model to give me an active manager's return. However I was hoping diversified indexes would have near zero alpha. I was hoping the FF model would be accurate enough to sniff out poor indexes or those with hidden costs, but its just too inaccurate. I currently see a 2% negative alpha on an index fund and it gives me pause; maybe it shouldn't...
And as Ron Scott notes, the only fund that really hasn't had much alpha (positive or negative) has been the total stock market, but this is only true for the FF 3 and 4 factor models. The wonky FF 5 factor model shows the total stock market with a 0.29% alpha for the past 25 years. Perhaps the expense ratio and trading costs play a role here, but I think the model is slightly off. Thanks for your thoughts guys.
Re: The negative alpha of the Value Factor
Totally ignored the "tstat". Besides actively managed funds such as VPMAX (which barely squeaks by with a tstat of 2.17) , the positive or negative alpha for pretty much any index fund put into the model (even VFINX since 1976) is not statistically significant.
Re: The negative alpha of the Value Factor
Recall that HML is calculated as 1/2 (Small Value + Big Value)  1/2 (Small Growth + Big Growth), so there is an outsized influence of the small value and small growth portfolios.
If small value outperforms relative to what is expected based on its value and size loads over a certain period of time (that is, if there's an interaction effect working to make small value better), that gives small value positive alpha. Which means that large value stocks would have positive exposure to the value factor but negative alpha.
Also keep in mind part of the value premium calculated historically is just the differential return above small growth, including the kinds of stocks with no earnings or very low liquidity that might not even make it into index funds that would be used. The small growth figures in the Fama/French formulation are pretty horrifying.
As you've noticed, some of these alphas may not be significant anyway.
If small value outperforms relative to what is expected based on its value and size loads over a certain period of time (that is, if there's an interaction effect working to make small value better), that gives small value positive alpha. Which means that large value stocks would have positive exposure to the value factor but negative alpha.
Also keep in mind part of the value premium calculated historically is just the differential return above small growth, including the kinds of stocks with no earnings or very low liquidity that might not even make it into index funds that would be used. The small growth figures in the Fama/French formulation are pretty horrifying.
As you've noticed, some of these alphas may not be significant anyway.
Re: The negative alpha of the Value Factor
Ah, the formula explains the negative alpha of Large Value (small value return was much higher than large value) and the positive alpha of Large Growth (small growth return was much lower than large growth). Thank you for pointing it out!lack_ey wrote: ↑Thu Nov 23, 2017 10:42 pmRecall that HML is calculated as 1/2 (Small Value + Big Value)  1/2 (Small Growth + Big Growth), so there is an outsized influence of the small value and small growth portfolios.
If small value outperforms relative to what is expected based on its value and size loads over a certain period of time (that is, if there's an interaction effect working to make small value better), that gives small value positive alpha. Which means that large value stocks would have positive exposure to the value factor but negative alpha.
Also keep in mind part of the value premium calculated historically is just the differential return above small growth, including the kinds of stocks with no earnings or very low liquidity that might not even make it into index funds that would be used. The small growth figures in the Fama/French formulation are pretty horrifying.
As you've noticed, some of these alphas may not be significant anyway.
This leads to another question. How are the returns for "Big Value" "Small Value" calculated? If the returns are valueweighted, the smallest microcaps will have higher book value because they are more risky  and therefore be overrepresented in calculating returns for the value factor. These distressed stocks should have substantially higher returns. Could this cause the value factor return to be deceptively high, resulting in negative alpha for even small value funds? Could valueweighting also increase transaction costs which are not accounted for?
Re: The negative alpha of the Value Factor
Answering my own question: valueweighted is capitalization weighted (see https://alphaarchitect.com/2017/07/24/m ... gs.5VPa1Yo). Therefore FF small value is a capitalization weighted index which could be achieved with reasonable portfolio turnover.