Vanguard Value Factor Fund factor regression surprise

Discuss all general (i.e. non-personal) investing questions and issues, investing news, and theory.
Post Reply
Topic Author
Kelly
Posts: 498
Joined: Sun Nov 18, 2012 6:39 am

Vanguard Value Factor Fund factor regression surprise

Post by Kelly »

Hi All

I'm trying to learn more about factors using portfolio visualizer.

Vanguard Factor (VFVA) has this M* style box

43 8 0
15 5 0
29 12 0

So I would expect a factor clone of LV and SV. Instead, it's 96% SV 4% MV: https://tinyurl.com/mkfhuufb

Why doesn't portfolio visualizer return LV + SV?

Thanks for any insight!

Kelly
muffins14
Posts: 5529
Joined: Wed Oct 26, 2016 4:14 am
Location: New York

Re: Vanguard Value Factor Fund factor regression surprise

Post by muffins14 »

I’ve never looked at this before, but It looks like you put in the funds available for the “clone”

You can probably get those factor loads with many possible combinations. For example just use IJS (small value) and VTV (large value) et voila, you get the LV + SV you’re looking for. There’s nothing special about it, I think. Perhaps it just outputs one out of the many possible solutions
Crom laughs at your Four Winds
Fryxell
Posts: 183
Joined: Sat Mar 02, 2013 9:08 pm
Location: Dallas

Re: Vanguard Value Factor Fund factor regression surprise

Post by Fryxell »

You’re using the wrong factor model. You want to make sure you include momentum and quality, because the Vanguard multi factor fund targets those. You’ll also want to include funds that target those factors.

That said, after doing that, the results aren’t all that different:

https://www.portfoliovisualizer.com/ma ... TAT_WEIGHT
Fryxell
Posts: 183
Joined: Sat Mar 02, 2013 9:08 pm
Location: Dallas

Re: Vanguard Value Factor Fund factor regression surprise

Post by Fryxell »

It’s important to keep in mind that many funds you might use as clone assets are themselves multi factor, which will obfuscate what you’re trying to do. For instance, I used VIOV, but this one has decent loading on value, size, quality anyway. You could try to use clone funds that are “purer,” such as VB for the size factor.

It’s usually best to just run the regression on the vanguard multifactor fund itself. This cloning thing is more of an entertaining curiosity. Or if you’re trying to clone an active fund, or decide if an apparently complex strategy can be simplified.
Northern Flicker
Posts: 15371
Joined: Fri Apr 10, 2015 12:29 am

Re: Vanguard Value Factor Fund factor regression surprise

Post by Northern Flicker »

Kelly wrote: Tue May 11, 2021 1:49 pm Hi All

I'm trying to learn more about factors using portfolio visualizer.

Vanguard Factor (VFVA) has this M* style box

43 8 0
15 5 0
29 12 0

So I would expect a factor clone of LV and SV. Instead, it's 96% SV 4% MV: https://tinyurl.com/mkfhuufb

Why doesn't portfolio visualizer return LV + SV?

Thanks for any insight!

Kelly
The clone is finding the best fit for the factor exposures, not an exact fit for the capitalization segments. The fund has substantial exposure to size and value, moderate exposure to quality, and no exposure to monentum in its regression:

https://www.portfoliovisualizer.com/fac ... sion=false

M* style boxes can be misleading because they may not draw the lines between large, mid, and small in alignment with your mental picture of the classes.
YRT70
Posts: 1289
Joined: Sat Apr 27, 2019 8:51 am

Re: Vanguard Value Factor Fund factor regression surprise

Post by YRT70 »

Fryxell wrote: Wed May 12, 2021 1:08 am You’re using the wrong factor model. You want to make sure you include momentum and quality, because the Vanguard multi factor fund targets those. You’ll also want to include funds that target those factors.
Just a heads up, afaik VFVA doesn't actively target momentum. The multifactor VFMF fund does.
Topic Author
Kelly
Posts: 498
Joined: Sun Nov 18, 2012 6:39 am

Re: Vanguard Value Factor Fund factor regression surprise

Post by Kelly »

Thanks for all the input. Here's where I get hung up: Bill Bernstein has stated that Vanguard's Value index has done a poor job capturing the value premium because it has too much large cap and not enough small cap where the premium mostly accrues.

This is only an academic lesson for myself. What fund would have a higher expected return and why? Vanguard's latest outlook calls for small cap returns equal to US equities and value 1% greater than US equities https://advisors.vanguard.com/insights/ ... cember2020

VIOV SMB .78 HML .63
VFVA SMB .38 HML .70

Would I expect VFVA with a higher value loading to have a higher return. Or, would VIOV have a higher return since it's much smaller ? Or am I confused in what these regression coefficients imply?

Thanks for any insight

Kelly
dbr
Posts: 46181
Joined: Sun Mar 04, 2007 8:50 am

Re: Vanguard Value Factor Fund factor regression surprise

Post by dbr »

Vanguard makes return forecasts using a proprietary Capital Markets Model which apparently is not the same as plugging factor loadings into a current Fama-French three factor regression.

I have no idea how one would find out the methodology and current assumptions made by the Vanguard model.

You could also look at Rick Ferri's 30 year forecast:

https://www.etf.com/sections/index-inve ... t-forecast

It might be the last time it appeared was 2015, but a 30 year forecast should not change much in six years, right?

Plenty of other forecasts exist. Note the Vanguard forecast is a ten year forecast, Ferri claims 30 years out, etc., etc.
Northern Flicker
Posts: 15371
Joined: Fri Apr 10, 2015 12:29 am

Re: Vanguard Value Factor Fund factor regression surprise

Post by Northern Flicker »

Kelly wrote: Wed May 12, 2021 8:29 am Thanks for all the input. Here's where I get hung up: Bill Bernstein has stated that Vanguard's Value index has done a poor job capturing the value premium because it has too much large cap and not enough small cap where the premium mostly accrues.

This is only an academic lesson for myself. What fund would have a higher expected return and why? Vanguard's latest outlook calls for small cap returns equal to US equities and value 1% greater than US equities https://advisors.vanguard.com/insights/ ... cember2020

VIOV SMB .78 HML .63
VFVA SMB .38 HML .70

Would I expect VFVA with a higher value loading to have a higher return. Or, would VIOV have a higher return since it's much smaller ? Or am I confused in what these regression coefficients imply?

Thanks for any insight

Kelly
I assume that Bill Bernstein likely was referring to VTV with his comment.

The regression coefficients are the coefficients that result from the the best fit of the linear model chosen to the data in the historical sample used.
Fryxell
Posts: 183
Joined: Sat Mar 02, 2013 9:08 pm
Location: Dallas

Re: Vanguard Value Factor Fund factor regression surprise

Post by Fryxell »

Kelly wrote: Wed May 12, 2021 8:29 am Thanks for all the input. Here's where I get hung up: Bill Bernstein has stated that Vanguard's Value index has done a poor job capturing the value premium because it has too much large cap and not enough small cap where the premium mostly accrues.

This is only an academic lesson for myself. What fund would have a higher expected return and why? Vanguard's latest outlook calls for small cap returns equal to US equities and value 1% greater than US equities https://advisors.vanguard.com/insights/ ... cember2020

VIOV SMB .78 HML .63
VFVA SMB .38 HML .70

Would I expect VFVA with a higher value loading to have a higher return. Or, would VIOV have a higher return since it's much smaller ? Or am I confused in what these regression coefficients imply?

Thanks for any insight

Kelly
Well, you’re running a monthly regression with data starting in 2018. So I wouldn’t take those loadings too seriously—not a lot of data points. You can try a daily regression, which can have its issues, but is helpful to look at in this situation:

https://www.portfoliovisualizer.com/fa ... sion=false

So now there is no difference in HML and looks like VIOV may also have a higher quality load and less negative momentum. I wouldn’t take the higher negative momentum for VFVA too seriously, until we see more data—that could be a fluke even with daily data (even if it comes out with low p-value).

You could also try the AQR factors. I don’t like them as much because I don’t fully understand them and sometimes they give weird results. But here goes:

https://www.portfoliovisualizer.com/fa ... sion=false

That seems to validate that VIOV loads more on quality. But suggests perhaps VFVA loads more on value.

Another thing you can do is look at the rolling regression to see if the factor loads are stable over time. The more stable they are, the less likely to be a fluke.

Which one can we expect to perform better? That’s a judgment call. Depends on your view on factor premiums. We can look up historical premiums for each factor, but they may do differently in the future. Going forward my guess is that value will probably do well.

It may come down to your view on size and quality. Do you believe in the size factor? If so, probably go with VIOV. Do you believe in quality factor? If so, that might give VIOV an edge if all else is equal (vanguard doesn’t really screen for quality in its flagship index funds).

Other things to keep in mind: If you go with VIOV, do you want to be all in small caps? VIOV also has a lot of REITS, which don’t work well with traditional factor models (they’re valued differently).

As for Bernstein, yes the value factor is stronger in small as a general statement. But I’m not sure that’s as true after controlling for quality. In any event, that doesn’t mean the vanguard large value funds don’t capture it as well—there’s just less to capture there.
Fryxell
Posts: 183
Joined: Sat Mar 02, 2013 9:08 pm
Location: Dallas

Re: Vanguard Value Factor Fund factor regression surprise

Post by Fryxell »

dbr wrote: Wed May 12, 2021 9:39 am Vanguard makes return forecasts using a proprietary Capital Markets Model which apparently is not the same as plugging factor loadings into a current Fama-French three factor regression.

I have no idea how one would find out the methodology and current assumptions made by the Vanguard model.
Good point. Most value metrics correlate with each other, so the regressions should still work broadly speaking. But would be interesting if they were using something like P/Earnings instead of P/Book. That would throw off a bit the regressions in portfolio visualizer. You’d get more drift in the VFVA value factor when regressing it with P/B.
Topic Author
Kelly
Posts: 498
Joined: Sun Nov 18, 2012 6:39 am

Re: Vanguard Value Factor Fund factor regression surprise

Post by Kelly »

Fryxell wrote: Thu May 13, 2021 1:08 am
Kelly wrote: Wed May 12, 2021 8:29 am Thanks for all the input. Here's where I get hung up: Bill Bernstein has stated that Vanguard's Value index has done a poor job capturing the value premium because it has too much large cap and not enough small cap where the premium mostly accrues.

This is only an academic lesson for myself. What fund would have a higher expected return and why? Vanguard's latest outlook calls for small cap returns equal to US equities and value 1% greater than US equities https://advisors.vanguard.com/insights/ ... cember2020

VIOV SMB .78 HML .63
VFVA SMB .38 HML .70

Would I expect VFVA with a higher value loading to have a higher return. Or, would VIOV have a higher return since it's much smaller ? Or am I confused in what these regression coefficients imply?

Thanks for any insight

Kelly
Well, you’re running a monthly regression with data starting in 2018. So I wouldn’t take those loadings too seriously—not a lot of data points. You can try a daily regression, which can have its issues, but is helpful to look at in this situation:

https://www.portfoliovisualizer.com/fa ... sion=false

So now there is no difference in HML and looks like VIOV may also have a higher quality load and less negative momentum. I wouldn’t take the higher negative momentum for VFVA too seriously, until we see more data—that could be a fluke even with daily data (even if it comes out with low p-value).

You could also try the AQR factors. I don’t like them as much because I don’t fully understand them and sometimes they give weird results. But here goes:

https://www.portfoliovisualizer.com/fa ... sion=false

That seems to validate that VIOV loads more on quality. But suggests perhaps VFVA loads more on value.

Another thing you can do is look at the rolling regression to see if the factor loads are stable over time. The more stable they are, the less likely to be a fluke.

Which one can we expect to perform better? That’s a judgment call. Depends on your view on factor premiums. We can look up historical premiums for each factor, but they may do differently in the future. Going forward my guess is that value will probably do well.

It may come down to your view on size and quality. Do you believe in the size factor? If so, probably go with VIOV. Do you believe in quality factor? If so, that might give VIOV an edge if all else is equal (vanguard doesn’t really screen for quality in its flagship index funds).

Other things to keep in mind: If you go with VIOV, do you want to be all in small caps? VIOV also has a lot of REITS, which don’t work well with traditional factor models (they’re valued differently).

As for Bernstein, yes the value factor is stronger in small as a general statement. But I’m not sure that’s as true after controlling for quality. In any event, that doesn’t mean the vanguard large value funds don’t capture it as well—there’s just less to capture there.
Very helpful! Another point of my confusion: Is it true that HML is not size dependent? It's just top 1/3 B/M minus bottom third? The historical data shows that SCV bettered LCV. Rereading ( and barley understanding) FF '92 paper, high B/M firms got there by a decline in market cap.

Your PV link shows that both VIOV and VFVA have a HML of 0.54. Can we say that high HML of a small firm has a higher expected return than larger firm with the same HML? Or, is it simply the SMB portion independent of HML?
Northern Flicker
Posts: 15371
Joined: Fri Apr 10, 2015 12:29 am

Re: Vanguard Value Factor Fund factor regression surprise

Post by Northern Flicker »

Kelly wrote: The historical data shows that SCV bettered LCV.
Correction: the data from one particular historical sample showed this. There is far too much sample bias in historical samples to assume any generalizability of returns.

Factor regressions are useful to compare the factor exposures of different portfolios by regressing them over the same time periods. Neither the models nor the fitting of data to them are accurate enough to offer much predictive value other than things like if the size and value premium turn out to be strong (or weak) then a fund with exposure to those factors will have a favorable (or mediocre) outcome.

The AQR factors are useful if you want size, value, quality, and momentum in a single regression.
Fryxell
Posts: 183
Joined: Sat Mar 02, 2013 9:08 pm
Location: Dallas

Re: Vanguard Value Factor Fund factor regression surprise

Post by Fryxell »

Kelly wrote: Thu May 13, 2021 9:00 am
Very helpful! Another point of my confusion: Is it true that HML is not size dependent? It's just top 1/3 B/M minus bottom third? The historical data shows that SCV bettered LCV. Rereading ( and barley understanding) FF '92 paper, high B/M firms got there by a decline in market cap.

Your PV link shows that both VIOV and VFVA have a HML of 0.54. Can we say that high HML of a small firm has a higher expected return than larger firm with the same HML? Or, is it simply the SMB portion independent of HML?
In the Fama/French model, yes, it is just the 1/3 minus the other 1/3 of the universe of stocks. The factor model assumes that the relationship is linear. But in reality it is not linear. We do know that factor effects vary by size and may have other interaction effects. You could in principle create your own factor model that captured the interaction effect by creating a Size*Value factor for a regression, something like that...

We could say the high HML stocks have a higher expected return than large ones. But you gotta be careful with the statement, as the factor model may capture some of that in the Size factor regression.

But yeah you could say that. That seems to be due to the “black hole” of small growth, which is mostly due to extremely overpriced “lottery” stocks. If you screen those out, the magnitude of the value effect in small stocks goes down (one reason why in actual practice, many small growth funds haven’t lagged small value by much). So that’s why I sad that if you screen for quality, the magnitude of the effect decreases.
Wade Garrett
Posts: 235
Joined: Wed Jul 10, 2013 2:41 pm
Location: Jasper, MO/Double Deuce

Re: Vanguard Value Factor Fund factor regression surprise

Post by Wade Garrett »

Kelly wrote: Thu May 13, 2021 9:00 am Is it true that HML is not size dependent?
read this......
https://www.bogleheads.org/blog/2020/04 ... -25-years/

It's never as simple as picking the factor fund with the best single factor regression for the factor you're targeting. There are different styles of value investing (Buffett style, Graham style, etc.) for one thing. Different metrics (p/e, p/cf, p/s, p/bv, ebit and ebitda, etc) for measuring value. Other factors (quality/profitability, negative momentum, etc) can play a role and impact value. I suggest doing a lot of reading to get a full understanding before diving head first into factor investing.
"I'm not an inventor. I'm an improver. I see things that are wrong, and I improve them." - Larry David, Curb Your Enthusiasm
Wade Garrett
Posts: 235
Joined: Wed Jul 10, 2013 2:41 pm
Location: Jasper, MO/Double Deuce

Re: Vanguard Value Factor Fund factor regression surprise

Post by Wade Garrett »

Kelly wrote: Wed May 12, 2021 8:29 am Bill Bernstein has stated that Vanguard's Value index has done a poor job capturing the value premium
Do you know for sure Bernstein was referring to VFVA? VFVA is an active fund and doesn't really follow an index. Like Northern Flicker said, he may have been referring to VTV or another value index.
"I'm not an inventor. I'm an improver. I see things that are wrong, and I improve them." - Larry David, Curb Your Enthusiasm
Topic Author
Kelly
Posts: 498
Joined: Sun Nov 18, 2012 6:39 am

Re: Vanguard Value Factor Fund factor regression surprise

Post by Kelly »

Wade Garrett wrote: Sun May 16, 2021 9:55 am
Kelly wrote: Wed May 12, 2021 8:29 am Bill Bernstein has stated that Vanguard's Value index has done a poor job capturing the value premium
Do you know for sure Bernstein was referring to VFVA? VFVA is an active fund and doesn't really follow an index. Like Northern Flicker said, he may have been referring to VTV or another value index.
Bernstein had never referred to VFVA that I know of. He has said that VTV has too large a market cap to capture the value premium. In Rational Expectations he replaced VTV (that was in Four Pillars) with the Vgd mid value index saying it's FF factors were closer to the DFA large value index.

I was trying to recreate his portfolios using less funds. He used both large and small cap values funds in his books while VFVA appears fit between and may take the place of both. VFVA appeared after his books.
Post Reply